CA2096425C - Methods for speech quantization and error correction - Google Patents

Abstract

The redundancy contained within the spectral amplitudes is reduced, and as a result the quantization of the spectral amplitudes is improved. The prediction of the spectral amplitudes of the current segment from the spectral amplitudes of the previous is adjusted to account for any change in the fundamental frequency between the two segments. The spectral amplitudes prediction residuals are divided into a fixed number of blocks each containing approximately the same number of elements. A prediction residual block average (PRBA) vector is formed; each element of the PRBA is equal to the average of the prediction residuals within one of the blocks The PRBA vector is vector quantized, or it is transformed with a Discrete Cosine Transform (DCT) and scalar quantized. The perceived effect of bit errors is reduced by smoothing the voiced/unvoiced decisions. An estimate of the error rate is made by locally averaging the number of correctable bit errors within each segment. If the estimate of the error rate is greater than a threshold, then high energy spectral amplitudes are declared voiced.

Description

Methods for Speech Quantization and Error Correction This invention relates to methods for quantizing speech and for preserving the quality-of speech during the presence of bit errors.
Relevant publications include: J. L. Flanagan, Speech Analysis, Synthesis and Per ception, Springer-Verlag, 1972, pp. 378-386, (discusses phase vocoder -frequency based speech analysis-synthesis system); GZuatieri, et al., "Speech Transformations Based on a Sinusoidal Representation", IEEE TASSP, Vol, ASSP34. No. 6, Dec.
1986, pp. 14-19-1986, (discusses analysis-synthesis technique based on a sinusoidal represen tation); Griffin, "Multiband Excitation Vocoder", Ph.D. Thesis, M.LT, 1987, (dis-cusses an 8000 bps Multi-Hand Excitation speech coder); Grif$n, et al., ''A
High GZual-ity 9.6 kbps Speech Coding System", Proc. ICASSP 86, pp. 125-128, Tokyo, Japan, April 13-20, 1986, (discusses a 9600 bps Multi-Band Excitation speech coder);Grif&n, et al., "A New Model-Based Speech Analysis/Synthesis System", Proc. ICASSP
~.~. p~. .~1.'3-~51F. Tampa. FL.. March ?6-29. 190. (discuses ~Iulti-Band Excitation speech model); Hardwick, "A 4.~ kbps Vlulti-Band Excitation Speech Coder", S.VI.
Thesis. VI.LT, May 1988, (discusses a 4800 bps Multi-Band Excitation speech coder);
VlcAuiay et al.. "Mid-Rate Coding Hased on a Sinusoidal Representation of Speech".
Proc. ICASSP 85, pp. 945-948. Tampa, FL., March 26-29, 1985, (discusses speech ?0 coding based on a sinusoidal representation); Campbell et al., "The Vow 4800 bps Voice Coding Standard'', A'Iil Speech Tech Conference, rov. 1989. (discusses er-ror correction in low rate speech coders); Campbell et al., "CELP Coding for Land h'Iobile Radio Applications", Proc. ICASSP 90. pp. 465-468, Albequerque, NM.
April 3-6, 1990, (discusses error correction in low rate speech coders);
Levesque et ~5 al., Error-Control Techniques for Digital Communication, Wiley, 1985, pp.
157-170, (discusses error correction in general); Jayant et al., Digital Coding of Waveforms, Prentice-Hall, 1984 (discusses quantization in general); Makhoul, et.al.
''Vector GZuantization in Speech Coding', Proc. IEEE, 1985, pp. 1551-1588 (discusses vector SUBSTITUTE SHEET

23. The method of cl quantization in general); Jayant et al., "Adaptive Postfiltering of 16 kb/s-ADPCM Speech", Proc. ICASSP 86, pp.
829-832, Tokyo, Japan, April 13-20, 1986, (discusses adaptive postfiltering of speech).
The problem of speech coding (compressing speech into a small number of bits) has a large number of applications, and as a result has received considerable attention in the literature. One class of speech coders (vocoders) which have been extensively studied and used in practice is based on an underlying model of speech.
Examples from this class of vocoders include linear prediction vocoders, homomorphic vocoders, and channel vocoders. In these vocoders, speech is modeled on a short-time basis as the response of a linear system excited by a periodic impulse train for voiced sounds or random noise for unvoiced sounds. For this class of vocoders, speech is analyzed by first segmenting speech using a window such as a Hamming window. Then, for each segment of speech, the excitation parameters and system parameters are estimated and quantized. The excitation parameters consist of the voiced/unvoiced decision and the pitch period. The system parameters consist of the spectral envelope or the impulse response of the system. In order to reconstruct speech, the quantized excitation parameters are used to synthesize an excitation signal consisting of a periodic impulse train in voiced regions or random noise in unvoiced regions. This excitation signal is then filtered using the quantized system parameters.
Even though vocoders based on this underlying speech model have been quite successful in producing 2a intelligible speech, they have not been successful in producing high-quality speech. As a consequence, they have not been widely used for high-quality speech coding. The poor quality of the reconstructed speech is in part due to the inaccurate estimation of the model parameters and in part due to limitations in the speech model.
A new speech model, referred to as the Multi-Band Excitation (MBE) speech WO 92/10830 ~ ~ ~ ~ I~ ~ ~ PCT/L1S91/09135 model, was developed by Griffin and Lim in 19S4. Speech coders based on this new speech model were developed by GrifF~n and Lim in 1986. and they were shown to be capable of producing high quality speech at rates above $000 bps (bits per second).
Subsequent work by Hardwick and Lim produced a 4800 bps ~fBE speech coder which p was also capable of producing high quality speech. This 4800 bps speech coder used more sophisticated quantization techniques to achieve similar quality at -1ti00 bps that earlier MBE speech coders had achieved at 8000 bps.
The 4800 bps \1BE speech coder used a RtBE analysis/synthesis system to esti mate the MBE speech model parameters and to synthesize speech from the estimated VIBE speech model parameters. A discrete speech signal, denoted by s(n), is obtained by sampling an analog speech signal. This is typically done at an 8 kHz.
sampling rate, although other sampling rates can easily be accommodated through a straight-forward change in the various system parameters. The system divides the discrete speech signal into small overlapping segments or segments by multiplying s(n) with a window- wf y ysuch as a Hamming window or a Ivaiser window ~ to obtain a windowed signal sw(n). Each speech segment is then analyzed to obtain a set of MBE
speech model parameters which characterize that segment. The MBE speech model param-eters consist of a fundamental frequency, which is equivalent to the pitch period, a set of voiced/unvoiced decisions, a set of spectral amplitudes, and optionally a set of spectral phases. These model parameters are then quantized using a fixed number of bits for each segment. The resulting bits can then be used to reconstruct the speech signal, by first reconstructing the VIBE model parameters from the bits and then synthesizing the speech from the model parameters. A block diagram of a typical MBE speech coder is shown in Figure 1.
?5 The 4800 bps 1~'IBE speech coder required the use of a sophisticated technique to quantize the spectral amplitudes. For each speech segment the number of bits which could be used to quantize the spectral amplitudes varied between 50 and 125 bits. In SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ PCT/US91/09135 addition the number of spectral amplitudes for each segment varies between 9 and 60. :1 quantization method was devised which could efficiently represent all of the spectral amplitudes with the number of bits available for each segment.
.although this spectral amplitude quantization method was designed for use in an ~iBE
speech coder the quantization techniques are equally useful in a number of different speech coding methods, such as the Sinusoidal Transform Coder and the Harmonic Coder .
For a particular speech segment, L denotes the number of spectral amplitudes in that segment. The value of L is derived from the fundamental frequency, w°, according to the relationship, io ~ _ ~~~- + .?~JJ (1) where 0 < ,3 < 1.0 determines the speech bandwidth relative to half the sampling rate. The function la~, referred to in Equation (1), is equal to the largest integer less than or equal to a. The L spectral amplitudes are denoted by M~ for 1 < !
< L, where :111 is the lowest frequency spectral amplitude and :11~ is the highest frequency 15 spectral amplitude.
The spectral amplitudes for the current speech segment are quantized by first calculating a set of prediction residuals which indicate the amount the spectral am-plitudes have changed between the current speech segment and the previous speech segment. If L° denotes the number of spectral amplitudes in the current speech 20 segment and L-1 denotes the number of spectral amplitudes in the previous speech segment, then the prediction residuals, T~ for 1 < ! < L° are given by, logz 111° - 7 ~ M~ ' if I < L'' ( ) loge :'l1° - ~ - :1~1L I, otherwise where .'~T° denotes the spectral amplitudes of the current speech segment and ~'~f~'' denotes the quantized spectral amplitudes of the previous speech segment. The con-stant 7 is typically equal to ."r , however any value in the range 0 < 7 < 1 can be used.
SUBSTITUTE SHEET

WO 92110830 ~ ~ ~ ~ A s~ ~ PCT/h'S91/09135 'The prediction residuals are divided into blocks of li elements, where the value ~f li is typically in the range -~ < li < 12. If L is not evenly divisible by li , then the highest frequency block will contain less than li elements. This is shown in Figure 2 for L = 3-f and li = 8.
Each of the prediction residual blocks is then transformed using a Discrete Cosine Transform (DCT) defined by, .~(k) _ ~ ~ ~(J)cos~~rk( ~ 2)~ (3) -o The length of the transform for each block, .1, is equal to the number of elements in the block. Therefore, all but the highest frequency block are transformed with a DCT of length li , while the length of the DCT for the highest frequency block is less than or equal to li . Since the DCT is an invertible transform, the L DCT
coefficients completely specify the spectral amplitude prediction residuals for the current segment.
The total number of bits available for quantizing the spectral amplitudes is divided among the DC'T coefi~cients according to a bit allocation rule. This rule attempts to give more bits to the perceptually more important low-frequency blocks, than to the perceptually less important high-frequency blocks. In addition the bit allocation rule divides the bits within a block to the DCT coefficients according to their relative long-term variances. This approach matches the bit allocation with the perceptual characteristics of speech and with the quantization properties of the DCT.
Each DCT coefficient is quantized using the number of bits specified by the bit allocation rule. Typically, uniform quantization is used, however non-uniform or vector quantization can also be used. The step size for each quantizer is determined from the long-term variance of the DCT coefficients and from the number of bits used to quantize each coefficient. Table 1 shows the typical variation in the step size as a function of the number of bits, for a long-term variance equal to Qz.
Once each DCT coefficient has .been quantized using the number of bits specified by the bit allocation rule, the binary representation can be transmitted, stored, etc..
SUBSTITUTE SHEET

WO 92/10830 ~ ~ (~ ~ /~ ~ ~ PCT/US91/09135 1'urnber .Step of Bits Sire 1 1.20 2 .850 3 .650 4 .420 5 .280 6 .140 7 .070 8 .0350 9 .01750 10 .008 7 11 .004380 12 .002190 13 .OO110o 14 .0005500 15 .0002750 16 .0001380 Table l: Step Size of L~uiform Quantizer~
depending on the application. The spectral amplitudes can be reconstructed from the binary representation by first reconstructing the quantized DCT
coefficients for each block. performing the inverse DCT on each block, and then combining with the quantized spectral amplitudes of the previous segment using the inverse of Equation (2). The inverse DCT is given by, ~(~) _ ~ a(7)~x(7) cos(~~~ J Z
,.o where the length, J, for each block is chosen to be the number of elements in that block, a( j) is given by, a~J) = 1 if j = 0 (5) 2 otherwise One potential problem with the 4800 bps MBE speech coder is that the perceived quality of the reconstructed speech may be significantly reduced if bit errors are added SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~~ !~ ~ ~ PCT/LlS9l/09135 _ ; _ to the binary representation of the \IBE model parau~eters. Since bit errors exist in many speech coder applications, a robust speech coder must be able to correct, detect and/or tolerate bit errors. One technique which has been found to be very successful is to use error correction codes in the binary representation of the model parameters.
Error correction codes allow infrequent bit errors to be corrected. and they allow the system to estimate the error rate. The estimate of the error rate can then be used to adaptively process the model parameters to reduce the effect of any remaining bit errors. Typically, the error rate is estimated by counting the number of errors corrected (or detected) by the error correction codes in the current segment, and then using this information to update the current estimate of error rate. For example if each segment contains a (23,12) Golay code which can correct three errors out of the ?3 bits, and eT denotes the number of errors (0-3) which were corrected in the current segment, then the current estimate of the error rate, ER, is updated according to:
ER=.3eR-ail-.3)~3 f6) where r3 is a constant in the range 0 < ;3 < 1 which controls the adaptability of ER.
When error correction codes or error detection codes are used, the bits representing the speech model parameters are converted to another set of bits which are more robust to bit errors. The use of error correction or detection codes typically increases the number of bits which must be transmitted or stored. The number of extra bits ?0 which must be transmitted is usually related to the robustness of the error correction or detection code. In most applications, it is desirable to minimize the total number of bits which are transmitted or stored. In this case the error correction or detection codes must be selected to maximize the overall system performance.
Another problem in this class of speech coding systems is that limitations in ~5 the estimation of the speech model parameters may cause quality degradation in the synthesized speech. Subsequent quantization of the model parameters induces further degradation. This degradation can take the form of reverberant or muffled SUBSTITV1'E SHEET

WO 92/10830 ~ ~ ~ ~ ~ ~ PCT/US91/09135 _ g _ quality to the synthesized speech. In addition background noise or other artifacts rnay be present which did not exist in the orignal speech. This form of degradation occurs even if no bit errors are present in the speech data, however bit errors can make this problem worse. Typically speech coding systems attempt to optimize the parameter estimators and parameter quantizers to minimize this form of degradation.
Other systems attempt to reduce the degradations by post-filtering. In post-filtering the output speech is filtered in the time domain with an adaptive all-pole filter to sharpen the format peaks. This method does not allow fine control over the spectral enhancement process and it is computationally expensive and inefficient for frequency l0 domain speech coders.
The invention described herein applies to many different speech coding methods, which include but are not limited to linear predictive speech coders, channel vocoders, homomorphic vocoders, sinusoidal transform coders, mufti-band excitation speech coders and improved multiband excitation (IMBE) speech coders. For the purpose of describing this invention in detail. we use the 6.-~ kbps 1\I13E speech coder which has recently been standardized as part of the INMARSAT-M (International Marine Satellite Organization) satellite communication system. This coder uses a robust speech model which is referred to as the Mufti-Band Excitation (MBE) speech model.
Efficient methods for quantizing the MBE model parameters have been developed.
These methods are capable of quantizing the model parameters at virtually any bit rate above 2 kbps. The 6.4 kbps IMBE speech coder used in the INMARSAT-M
satellite communication system uses a 50 Hz frame rate. Therefore 128 bits are available per frame. Of these 128 bits, 45 bits are reserved for forward error correction.
The remaining 83 bits per frame are used to quantize the MBE model parameters, which consist of a fundamental frequency wo, a set of V/UV decisions vk for 1 < k <
li , and a set of spectral amplitudes tl-1, for 1 < 1 < L. The values of K and L vary depending on the fundamental frequency of each frame. The 83 available bits are SUBSTITUTE SHEET

WO 9Z/10830 PCT/US91l09135 -~~96~~~
_ 4 _ Parameter :'~'um6er of Bits Fundamental FrequencyS

Voiced/Unvoiced DecisionsIi Spectral Amplitudes 75 - Ii Table 2: Bit Allocation Among Model Parameters divided among the model parameters as shown in Table 2.
The fundamental frequency is quantized by first converting it to its equivalent pitch period using Equation (7).
Po - 2~r ~7) The value of Po is typically restricted to the range 20 < Po < 120 assuming an 8 kHz sampling rate. In the 6.4 kbps IMBE system this parameter is uniformly quantized using 8 bits and a step size of ..5. This corresponds to a pitch accuracy of one half sample.
The Ii V/UV decisions are binary values. Therefore they can be encoded using a single bit per decision. The 6.4 kbps system uses a maximum of 12 decisions, and the width of each frequency band is equal to 3~0. The width of the highest frequency band is adjusted to include frequencies up to 3.8 kHz.
The spectral amplitudes are quantized by forming a set of prediction residuals.
2o Each prediction residual is the difference between the logarithm of the spectral ampli-tude for the current frame and the logarithm of the spectral amplitude representing the same frequency in the previous speech frame. The spectral amplitude prediction residuals are then divided into six blocks each containing approximately the same number of prediction residuals. Each of the six blocks is then transformed with a Discrete Cosine Transform (DCT) and the D.C. coefficients from each of the six blocks are combined into a 6 element Prediction Residual Block Average (PRBA) SllBSTiTUTE SHEET

WO 92/ 10830 ~ ~ ~ ~ ~ ~ PCT/US9l /09135 s _ l vector. The mean is subtracted from the PRBA vector and quantized using a 6 bit non-uniforcn quantizer. The zero-mean PRBa vector is then vector quantized using a 10 bit vector quantizer. The 10 bit PRBA codebook was designed using a k-means clustering algorithm on a large training set consisting of zero-mean PRBA
vectors p from a variety of speech material. The higher-order DCT coefficients which are not included in the PRBA vector are quantized with scalar uniform quantizers using the 59 - li remaining bits. The bit allocation and quantizer step sizes are based upon the long-term variances of the higher order DCT coefficients.
There are several advantages to this quantization method. First, it provides very good fidelity using a small number of bits and it maintains this fidelity as L
varies over its range. In addition the computational requirements of this approach are well within the limits required for real-time implementation using a single DSP
such as the AT&T DSP32C. Finally this quantization method separates the spectral amplitudes into a few components, such as the mean of the PRBA vector, which are sensitive to bit errors, and a large number of other components which are not very sensitive to bit errors. Forward error correction can then be used in an efficient manner by providing a high degree of protection for the few sensitive components and a lesser degree of protection for the remaining components. This is discussed in the next section.
In a first aspect, the invention features an improved method for forming the pre-dieted spectral amplitudes. They are based on interpolating the spectral amplitudes of a previous segment to estimate the spectral amplitudes in the previous segment at the frequencies of the current segment. This new method corrects for shifts in the frequencies of the spectral amplitudes between segments, with the result that the prediction residuals have a lower variance, and therefore can be quantized with less distortion for a given number of bits. In preferred embodiments, the frequencies of the spectral amplitudes are the fundamental frequency and multiples thereof.
In a second aspect, the invention features an improved method for dividing the suBST~Tu~ sHE~

WO 92/10830 ~ ~ ~ ~ ~ ~ ~ PCT/US91/09135 prediction residuals into blocks. Instead of fixing the length of each block and then dividing the prediction residuals into a variable number of blocks, the prediction residuals are divided into a predetermined number of blocks and the size of the blocks varies from segment to segment. In preferred embodiments, six (6) blocks are used in all segments: the number of prediction residuals in a lower frequency block is not larger that the number of prediction residuals in a higher frequency block; the difference between the number of elements in the highest frqeuency block and the number of elements in the lowest frequency block is less than or equal to one.
This new method more closely matches the characteristics of speech, and therefore it allows the prediction residuals to be quantized with less distortion for a given number of bits. In addition it can easily be used with vector quantization to further improve the quantization of the spectral amplitudes.
In a third aspect, the invention features an improved method for quantizing the prediction residuals. The prediction residuals are grouped into blocks, the average of the prediction residuals within each block is determined, the aaerages of all of the blocks are grouped into a prediction residual block average (PRBA) vector, and the PRBA vector is encoded. In preferred embodiments, the average of the prediction residuals is obtained by adding the spectral amplitude prediction residuals within the block and dividing by the number of prediction residuals within that block, or by computing the DCT of the spectral amplitude prediction residuals within a block and using the first coefficient of the DCT as the average. The PRBA vector is preferably encoded using one of two methods: ( 1 ) performing a transform such as the DCT on the PRBA vector and scalar quantizing the transform coefficients;
(2) vector quantizing the PRBA vector. Vlector quantization is preferably performed by determining the average of the PRBA vector, quantizing said average using scalar quantization, and quantizing the zero-mean PRBA vector using vector quantization with a zero-mean code-book. An advantage of this aspect of the invention is that it SUBSTITUTE SHEET

- 1? -allows the prediction residuals to be quantized with less distortion for a given number of bits.
In a fourth aspect, the invention features an unproved method for determining the voiced/unvoiced decisions in the presence of a high bit error rate. The bit error rate is estimated for a current speech segment and compared to a predetermined error-rate threshold, and the voiced/unvoiced decisions for spectral amplitudes above a predetermined energy threshold are all declared voiced for the current segment when the estimated bit error rate is above the error-rate threshold. This reduces the perceptual effect of bit errors. Distortions caused by switching from voiced to unvoiced are reduced.
In a fifth aspect, the invention features an improved method for error correction (or error detection) coding of the speech model parameters. The new method uses at least two types of error correction coding to code the quantized model parameters. A
first type of coding, which adds a greater number of additional bits than a second type 1~ of coding, is used for a group of parameters that is more sensitive to bit errors. The other type of error correction coding is used for a second group of parameters that is less sensitive to bit errors than the first. Compared to existing methods, the new method improves the quality of the synthesized speech in the presence of bit errors while reducing the amount of additional error correction or detection bits which must be added. In preferred embodiments, the different types of error correction include Golay codes and Hamming codes.
In a sixth aspect, the invention features a further method for improving the quality of synthesized speech in the presence of bit errors. The error rate is estimated from the error correction coding, and one or more model parameters from a previous segment are repeated in a current segment when the error rate for the parameters exceeds a predetermined level. In preferred embodiments, all of the model parameters are repeated.
SUBSTITUTE SHEET

In a seventh aspect, the invention features a new method for reducing the degradation caused by the estimation and quantization of the model parameters. This new method uses a frequency domain representation of the spectral envelope parameters to enhance regions of the spectrum which are perceptually important and to attenuate regions of the spectrum which are perceptually insignificant. The result is that degradation in the synthesized speech is reduced. A
smoothed spectral envelope of the segment is generated by smoothing the spectral envelope, and an enhanced spectral envelope is generated by increasing some frequency regions of the spectral envelope for which the spectral envelope has greater amplitude than the smoothed envelope and decreasing some frequency regions for which the spectral envelope has lesser amplitude than the smoothed envelope. In preferred embodiments, the smoothed spectral envelope is generated by estimating a low-order model (e. g. an all-pole model) from the spectral envelope. Compared to existing methods this new method is more computationally efficient for frequency domain speech coders. In addition this new method improves speech quality by removing the frequency domain constraints imposed by time-domain methods.
The invention may be summarized as a method of encoding speech wherein the speech is broken into segments, each of said segments representing one of a succession of time intervals and having a spectrum of frequencies, and for each segment the spectrum of frequencies is sampled at a set of frequencies to form a set of actual spectral amplitudes, with the frequencies at which the spectrum of frequencies is sampled generally differing from one segment to the next, and wherein the spectral amplitudes for at least one previous segment are used to produce a set of predicted 13a spectral amplitudes for a current segment, and wherein a set of prediction residuals for the current segment based on a difference between the actual spectral amplitudes for the current segment and the predicted spectral amplitudes for a current segment are used in subsequent encoding, characterized in that the prediction residuals for a segment are grouped into blocks, the prediction residuals within each block are determined, the averages of each of the blocks are grouped into a prediction residual block average (PRBA) vector, and the PRBA vector is encoded.
Other features and advantages of the invention will be apparent from the following description of preferred embodiments and from the claims.
Brief Description of the Drawings Figures 1-2 are diagrams showing prior art speech coding methods.
Figure 3 is a flow chart showing a preferred embodiment of the invention in which the spectral amplitudes are divided into a fixed number of blocks.
Figure 4 is a flow chart showing a preferred embodiment of the invention in which the spectral amplitude prediction accounts for any change in the fundamental frequency.
Figure 5 is a flow chart showing a preferred embodiment of the invention in which a prediction residual block average vector is formed.

WO 92/10830 ~ 9 6 I~ 'Z ~ PCT/US91/09135 Figure 6 is a flow chart showing a preferred embodiment of the invention in which the prediction residual block average vector is vector duantized Figure 7 is a flow chart showing a preferred embodiment of the invention in which the prediction residual block average vector is quantized with a DCT and scalar quantization.
Figure 8 is a flow chart showing a preferred embodiment of the invention encoder in which different error correction codes are used for different model parameter bits.
Figure 9 is a flow chart showing a preferred embodiment of the invention decoder in which different error correction codes are used for different model parameter bits.
l0 Figure 10 is a flow chart showing a preferred embodiment of the invention in which frequency domain spectral envelope parameter enhancement is depicted.
Description of Preferred Embodiments of the Invention In the prior art. the spectral amplitude prediction residuals were formed using Equation ('?). This method does not account for any change in the fundamental frequency between the previous segment and current segment. In order to account for the change in the fundamental frequency a new method has been developed which first interpolates the spectral amplitudes of the previous segment. This is typically done using linear interpolation, however various other forms of interpolation could also be used. Then the interpolated spectral amplitudes of the previous segment are resampled at the frequency points corresponding to the multiples of the fundamental frequency of the current segment. This combination of interpolation and resampling produces a set of predicted spectral amplitudes, which have ~~en corrected for any inter-segment change in the fundamental frequency.
Typically a fraction of the base two logarithm of the predicted spectral amplitudes is subtracted from the base two logarithm of the spectral amplitudes of the current segment. If linear interpolation is used to compute the predicted spectral amplitudes, SU~STtTUTE ~H~~

WO 92/10830 ~ /~ ~' ~ PCT/US91109135 then this can be expressed mathematically as:
T~ - loge .~l° - -; (( 1 - cat) Iog2 .11k' ~ ~, log .lf~_ Tli l~) where br is given bv, - -,o -,o ~r=;:~ol .l_ Two' .ll (9) where 7 is a constant subject to 0 < 7 < 1. Typically, ~ _ .r, however other values of 7 can also be used. For example 7 could be adaptively changed from segment to segment in order to improve performance. The parameters :ao and wo' in Equation (9) refer to the fundamental frequency of the current segment and the previous segment, respectively. In the case where the two fundamental frequencies are the same, the new method is identical to the old method. In other cases the new method produces a prediction residual with lower variance than the old method.
This allows the prediction residuals to be quantized with less distortion for a given number of bits.
In another aspect of the itwentiun a new method hay been developed to divide the spectral amplitude prediction residuals into blocks. In the old method the L
prediction residuals from the current segment were divided into blocks of li elements.
where li - S is a typical value. Using this method, the characteristics of each block were found to be significantly different for large and small values of L. This ~0 reduced the quantization efficiency, thereby increasing the distortion in the spectral amplitudes. In order to make the characteristics of each block more uniform, a new method was divised which divides the L prediction residuals into a fixed number of blocks. The length of each block is chosen such that all blocks within a segment have nearly the same length, and the sum of the lengths of all the blocks within a segment equal L. Typically the total number of prediction residuals is divided into 6 blocks, where the length of each block is equal to l s ~ . If L is not evenly divisible by 6 then the length of one or more higher frequency blocks is increased by one, such that all of the spectral magnitudes are included in one of the six blocks.
This new SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ PCT/L)S91/09135 - lb -method is shown in Figure -1 for the case where 6 blocks are used and L = 34.
In this new method the approximate percentage of the prediction residuals contained in each block is independent of L. This reduces the variation in the characteristics of each block, and it allows more efficient quantization of the prediction residuals.
The quantization of the prediction residuals can be further improved by forming a prediction residual block average (PRBA) vector. The length of the PRBA
vector is equal to the number of blocks in the current segment. The elements of this vector correspond to the average of the prediction residuals within each block. Since the first DCT coefficient is equal to the average (or D.C. value), the PRBA vector can be formed from the first DCT coefficient from each block. This is shown in Figure 5 for the case where 6 blocks are present in the current segment and L = 34. This process can be generalized by forming additional vectors from the second (or third, fourth, etc.) DCT coefficient from each block.
The elements of the PRBA vector are highly correlated. Therefore a number of methods can be used to improve the quantization of the spectral amplitudes.
One method which can be used to achieve very low distortion with a small number of bits is vector quantization. In this method a codebook is designed which contains a number of typical PRBA vectors. The PRBA vector for the current segment is compared against each of the codebook vectors, and the one with the lowest error is chosen as the quantized PRBA vector. The codebook index of the chosen vector is used to form the binary representation of the PRBA vector. A method for performing vector quantization of the PRBA vector has been developed which uses the cascade of a 6 bit non-uniform quantizer for the mean of the vector, and a 10 bit vector quantizer for the remaining information. This method is shown in Figure 6 for the case where the PRBA vector always contains 6 elements. Typical values for the 6 bit and 10 bit quantizers are given in the attached appendix.
An alternative method for quantizing the PRBA vector has also been developed.
SUBSTITUTE SHEET

WO 92!10830 ~ ~ ~ ~ ~ ~ ~~ PC'I'1US91/09135 - li -This method requires less computation and storage than the vector quantization method. In this method the PRBA vector is first transformed with a DGT as defined in Equation (3). The length of the DCT is equal to the number of elements in the PRBA ~~ector. The DCT coefficients are then quantized in a manner similar to that discussed in the prior art. First a bit allocation rule is used to distribute the total number of bits used to quantize the PRBA vector among the DCT
coefficients.
Scalar quantization (either uniform or non-uniform) is then used to quantize each DCT coefficient using the number of bits specified by the bit allocation rule.
This is shown in Figure r for the case where the PRBA vector always contains 6 elements.
Various other methods can be used to efficiently quantize the PRBA vector. For example other transforms such as the Discrete Fourier Transform, the Fast Fourier Transform, the Karhunen-Louve Transform could be used instead of the DCT. In addition vector quantization can be combined with the DCT or other transform.
The improvements derived from this aspect of the invention can be used with a wide variety of quantization methods.
In another aspect a new method for reducing the perceptual effect of bit errors has been developed. Error correction codes are used as in the prior art to correct infrequent bit errors and to provide an estimate of the error rate eR. The new method uses the estimate of the error rate to smooth the voiced/unvoiced decisions, in order to reduce the perceived effect of any remaining bit errors. This is done by first comparing the error rate against a threshold which signifies the rate at which the distortion from uncorrected bit errors in the voiced/unvoiced decisions is significant.
The exact value of this threshold depends on the amount of error correction applied to the voiced/unvoiced decisions, but a threshold value of .003 is typical if little error correction has been applied. If the estimated error rate, ER, is below this threshold then the voiced/unvoiced decisions are not perturbed. If eR is above this threshold SUBSTITUTE SHEET

~~~~~~~5 then every spectral amplitude for which Equation ( 10) is satisfied is declared voiced.
a;.2ss~sE>''' if .003 < ER < .p?
II/ ~ ~ra~r~3.29cR) (10) 1.-~1~(SE)~375 1f ER J
although Equation ( 10) assumes a threshold value of .003. this method can easily be modified to accommodate other thresholds. The parameter SE is a measure of the local average energy contained in the spectral amplitudes. This parameter is typically updated each segment according to:
.95 SE + Ø5 Ra if .93 SE + .05 Ro <_ 10000.0 SE = (11) 10000.0 otherwise where Ro is given by, L
Ra = ~ 1t1,2 ( 12 ) m The initial value of SE is set to an arbitrary initial value in the range 0 <
SE <
10000Ø The purpose of this parameter is to reduce the dependency of Equation j 10 on the average signal level. This ensures that the new method works as well for low level signals as it does for high level signals.
The specific forms of Equations (10), (11) and (12) and the constants contained within them can easily be modified, while maintaining the essential components of the new method. The main components of this new method are to first use an estimate of the error rate to determine whether the voiced/unvoiced decisions need to be smoothed. Then if smoothing is required, the voiced/unvoiced decisions are perturbed such that all high energy spectral amplitudes are declared voiced. This eliminates any high energy voiced to unvoiced or unvoiced to voiced transitions between segments.
and as a result it improves the perceived quality of the reconstructed speech in the presence of bit errors.
In our invention we divide the quantized speech model parameter bits into three or more different groups according to their sensitivity to bit errors, and then we SUBSTITUTE l6HEET

WO 92/10830 ; ~ ~ ~ ~ ~ ~ PCT/US91/09135 use different error correction or detection codes for each group. Typically the group of data bits which is determined to be most sensitive to bit errors is protected using very effective error correction codes. Less effective error correction or uetection codes.
which require fewer additional bits, are used to protect the less sensitive data bits.
This new method allows the amount of error correction or detection given to each group to be matched to its sensitivity to bit errors. Compared to the prior art, this method has the advantage that the degradation caused by bit errors is reduced and the number of bits required for forward error correction is also reduced.
The particular choice of error correction or detection codes which is used depends upon the bit error statistics of the transmission or storage medium and the desired bit rate. The most sensitive group of bits is typically protected with an effective error correction code such as a Hamming code, a BCH code, a Golay code or a Reerd-Solomon code. Less sensitive groups of data bits may use these codes or an error detection code. Finally the least sensitive groups may use error correction or detection codes or they may not use any form of error correction or detection. The w-ention is described herein using a particular choice of error correction and detection codes which was well suited to a 6.4 kbps MBE speech coder for satellite communications.
In the 6.~ kbps IV4BE speech coder, which was standardized for the I~'i~irIRSAT
VI satellite communciation system, the 45 bits per frame which are reserved for for ward error correction are divided among 23,12) Golay codes which can correct up to 3 errors, (15,11) Hamming codes which can correct single errors and parity bits. The six most significant bits from the fundamental frequency and the three most signif-icant bits from the mean of the PRBA vector are first combined with three parity check bits and then encoded in a (23,12) Golay code. A second Golay code is used to encode the three most significant bits from the PRBA vector and the nine most sensitive bits from the higher order DCT coefficients. All of the remaining bits except the seven least sensitive bits are then encoded into five (15,11) Hamming codes. The SUBSTITUTE SHE ,~~T

WO 92/ 10830 ~ PCT/ US91 /09135 ~~~6!~2~

seven least significant bits are not protected with error correction codes.
Prior to transmission the 1'?~ bits which represent a particular speech segment are interleaved such that at least five bits separate any two bits from the same code word.
This feature spreads the effect of short burst errors over several different codewords, thereby increasing the probability that the errors can be corrected.
At the decoder the received bits are passed through Golay and Hamming decoders which attempt to remove any bit errors from the data bits. The three parity check bits are checked and if no uncorrectable bit errors are detected then the received bits are used to reconstruct the MBE model parameters for the current frame.
Otherwise to if an uncorrectable bit error is detected then the received bits for the current frame are ignored and the model parameters from the previous frame are repeated for the current frame.
The use of frame repeats has been found to improve the perceptual quality of the speech when bit errors are present. The invention examines each frame of received bits and determines whether the current frame is likely to contain a large number of uncorrectable bit errors. One method used to detect uncorrectable bit errors is to check extra parity bits which are inserted in the data. The invention also determines whether a large burst of bits errors has been encountered by comparing the number of correctable bit errors with the local estimate of the error rate. If the number of correctable bit errors is substantially greater than the local estimate of the error rate then a frame repeat is performed. Additionally, the invention checks each frame for invalid bit sequences (i.e. groups of bits which the encoder never transmits).
If an invalid bit sequence is detected a frame repeat is performed.
The Golay and Hamming decoders also provide information on the number of correctable bit errors in the data. This information is used by the decoder to estimate the bit error rate. The estimate of the bit error rate is used to control adaptive smoothers which increase the perceived speech quality in the presence of uncorrectable ~~iBSTiTUTE ~~E~

WO 92/10830 2 ~ 9 G ~ l 5 PCT/US91/09135 bit errors. In addition the estimate of the error rate can he used to perform frame repeats in bad error environments.
This aspect of the invention can be used with soft-decision coding to further improve performance. Soft-decision decoding uses additional information on the like s lihood of each bit being in error to improve the error correction and detection capabil ities of many different codes. Since this additional information is often available from a demodulator in a digital communication system, it can provide improved robustness to bit errors without requiring additional bits for error protection.
The invention uses a new frequency domain parameter enhancement method which l0 improves the quality of synthesized speech. The invention first locates the percep-tually important regions of the speech spectrum. The invention then increases the amplitude of the perceptually important frequency regions relative to other frequency regions. The preferred method for performing frequency domain parameter enhance-ment is to smooth the spectral envelope to estimate the general shape of the spectrum.
15 The spectrum can Le smoothed by fitting a low-order model such as an all-pole model, a cepstral model, or a polynomial model to the spectral envelope. The smoothed spec-tral envelope is then compared against the unsmoothed spectral envelope and per-ceptually important spectral regions are identified as regions where the unsmoothed spectral envelope has greater energy than the smoothed spectral envelope.
Similarly 20 regions where the unsmoothed spectral envelope has less energy than the smoothed spectral envelope are identified as perceptually less important. Parameter enhance-ment is performed by increasing the amplitude of perceptually important frequency regions and decreasing the amplitude of perceptually less important frequency re-gions. This new enhancement method increases speech quality by eliminating or 25 reducing many of the artifacts which are introduced during the estimation and quan-tization of the speech parameters. In addition this new method improves the speech intelligibility by sharpening the perceptually important speech formants.
gUBSTtTUT~ SHEET

2~~~-~~~
In the MBE speech decoder a first-order all-pole model is fit to the spectral envelope for each frame. This is done by estimating the correlation parameters, Ro and R1 from the decoded model parameters according to the following equations, G
Ro = ~ .11,2 ( 13 ) m L, R1 =~~Il2cos(wol) (1.t) i= ~
where ~1~ for 1 < l < L are the decoded spectral amplitudes for the current frame, and wo is the decoded fundamental frequency for the current frame. The correlation parameters Ro and R1 can be used to estimate a first-order all-pole model.
This model is evaluated at the frequencies corresponding to the spectral amplitudes for the current frame (i.e. k ~ wo for 1 < l < L) and used to generate a set of weights G~', according to the following formula.
- ~ .96~r(Ro + R? - 2RaR1 cos(wo !)) ' for 1 < l < L ~ 1 ~l lt-; _ 11~ ~ ~ wORo(RO _ Ri) These weights indicate the ratio of the smoothed all-pole spectrum to the IMBE
spectral amplitudes. They are then used to individually control the amount of pa-rameter enhancement which is applied to each spectral amplitude. This relationship is expressed in the following equation, 1.2~M~ if W, > 1.2 ~LTi = for 1 < l < L ( 16) W~ ~ M~ otherwise where R'Ti for 1 < l < L are the enhanced spectral amplitudes for the current frame.
The enhanced spectral amplitudes are then used to perform speech synthesis.
The use of the enhanced model parameters improves speech quality relative to synthesis from the unenhanced model parameters.
Further description of the preferred embodiment is given in the attached Ap-pendix.
SUBSTITUTE SHfET

WO 92/10830 ~ ~:6 ' ~ ~ PCT/US91/09135 INMARSAT M Voice Codec ~Copyright, Digital Voice Systems Inc., 1991 Version 1.7 13 September 1991 Digital Voice Systems Inc. (DVSI, L'.S.A.) claims certain rights in the Improved Multi-Band Excitation voice coding algorithm described in this document and elsewhere in the I~':~tARS.~T Vi System Definition Lianual. DVSI is willing to grant a royalty-free license to ase the I~fBE voice coding algorithm strictly in connection with the I'~''~iARSAT M system on certain conditions of reciprocity. Details may be obtained from INMARSAT.
This document may contain errors in the description of the IMBE voice coding algo-rithm. An attempt will be made to correct these errors in future updates of this document.
DVSI acknowledges the '~iassa~chusetts Institute of Technology where the ~fulti-Band Excitation speech model was developed. In addition DVSI acknowledges the Rome A.ir Development Center of the United States Air Force which supported the real-time hardware used in the I:~'Li:~RSAT/AUSSAT voice coding evaluation.
SUBSTITUTE SHEET

WO 92/10830 ~' ~ ~ ~ ~ ~ ~ PCT/US9l/09135 Contents 1 Introduction 2 Multi-Band Excitation Speech Model 7 3 Speech Analysis 11 3.1 Pitch Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . 13 3.1.1 Determination of E(P) . . . . . . . . . . . . . . . . . . . . . . . . .

3.1.2 Pitch Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. 15 3.1.3 Look-Back Pitch TSracking . . . . . . . . . . . . . . . . . . . . . . .
. 16 3.1.4 Look-Ahead Pitch Tracking . . . . . . . . . . . . . . . . . . . . . . .

3.1.5 Pitch Refinement . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. 19 3.2 Voiced/Unvoiced Determination . . . . . . . . . . . . . . . . . . . . . .
. . . 21 3.3 Estimation of the Spectral Amplitudes . . . . . . . . . . . . . . . . . .
. . . 23 4 Parameter Encoding and Decoding 25 4.1 Fundamental Frequency Encoding and Decoding . . . 25 . . . . . . . . . . . .

4.2 ~oiced/Unvoiced Decisiaa Encoding and Decoding . 27 . . . . . . . . . . . . .

4.3 Spectral Amplitudes Encoding . . . . . . . . . . 27 . . . . . . . . . . . . . . .

4.3.I Encoding the PRBA Vector . . . . . . . . . . . 31 . . . . . . . . . . . .

4.3.2 Encoding the Higher Order DCT Coeffiaents . . . 32 . . . . . . . . . .

4.4 Spectral Amplitndes Decoding . . . . . . . . . . 34 . . . . . . . . . . . . . . .

4.4.1 Decoding the PRBA Vector . . . . . . . . . . . 36 . . . . . . . . . . . .

4.4.2 Decoding the higher Order DCT Coefficients . . 37 . . . . . . . . . . .

4.4.3 Spectral Amplitude Enhancement . . . . . . . . 37 . . . . . . . . . . .

Forward Error Correction and Hit Interleaving 39 5.1 Error Correction Encodiag . . . . . . . . . . . . . . . . . . . . . . . .
. . . . 39 5.2 Bit Interleaving . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . 42 5.3 Error Correction Decoding . . . . . . . . . . . . . . . . . . . . . . . .
. . . . 42 5.4 Adaptive Smoothing . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . 43 SUSS1'I?UTE SHEET

8 Parameter Encoding Example - 25 -4b '~ Speech Synt hesis 5I
..1 Speech Synthesis Notation . . . . . . . , , , . , . . . . . . . . . . . .
. . . , 51 7.2 Lnvoiced Speech Synthesis . . . . . . . . . . . . . . . . . . . . . . . .
. . . . 52 7.3 Voiced Speech Synthesis . . . . . . , , , , , , , , , . . , . . . . , . .
. . . . 54 8 Additional Notes S7 A Variable Initialization S8 B Initial Pitch Estimation Window 59 C Pitch Refinement Window el D FIR Low Pass Filter 83 E Mean Prediction Residual Quantizer Levels 84 F Prediction Residual Hlock Average Quantization Vectors 8S
G Spectral Amplitude Hit Allocation 9T
H Hit Frame Format 118 I Speech Synthesis Window 120 J Flow Charts 122 SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ PC1'/US9l/09135 List of Figures 1 Improved Multi-Band Excitation Speech Coder . . . 7 . . . . . . . . . . . . .

2 Comparison of Traditional and MBE Speech Models . 9 . . . . . . . . . . . .

3 IMBE Speech Analysis Algorithm . . . . . . . . . 11 . . . . . . . . . . . . . . .

4 High Pass Filter Frequency Response at 8 kHz. Sampling12 Rate . . . . . . .

Relationship between Speech Frames . . . . . . . 13 . . . . . . . . . . . . . . .

6 ndow Alignment . . . . . . . . . . . . . . . . . 14 . . . . . . . . . . . . . . .

Initial Pitch Estimation . . . . . . . . . . . . 16 . . . . . . . . . . . . . . . . .

8 Pitch Refinement . . . . . . . . . . . . . . . . 19 . . . . . . . . . . . . , . . . .

9 IMBE Voiced/LTnvoiced Determination . . . . . . . 21 . . . . . . . . . . . . . .

10IMBE Frequency Band Structure . . . . . . . . . . 23 . . . . . . . . . . . . . .

11IMBE Spectral Amplitude Estimation . . . . . . . 24 . . . . . . . . . . . . . .

12Fundamental Frequency Encoding and Decoding . . . 27 . . . . . . . . . . . .

13V/UV Decision Encoding and Decoding . . . . . . . 28 . . . . . . . . . . . . .

14Encoding of the Spectral Amplitudes . . . . . . . 28 . . . . . . . . . . . . . . .

15Prediction Residual Blocks for L = 34 . . . . . . 29 . . . . . . . . . . . . . . .

16Formation of Prediction Residual Block Average Vector30 . . . . . . . . . . .

17Decoding of the Spectral Amplitudes . . . . . . . 36 . . . . . . . . . . . . . . .

18Error Correction Encoding . . . . . . . . . . . . 39 . . . . . . . . . . . . . . . .

I9Format of ua and cd . . . . . . . . . . . . . . . 41 . . . . . . . . . . . . . . . .

20Format of ul and c~ . . . . . . . . . . . . . . . 41 . . . . . . . . . . . . . . . .

21IMBE Speech Synthesis . . . . . . . . . . . . . . 52 . . . . . . . . . . . . . . .

List of Tables I Bit Allocation Among Model Parameters . . . . . . . . . . . . . . . . . . .

2 Eight Bit Binary Representation . . . . . . . . . . . . . . . . . . . . . .
. . 26 3 Standard Deviation of PRBA Quantization Errors . . . . . . . . . . . . . .

4 Step Size of Uniform Quantizers . . . . . . . . . . . . . . . . . . . . . .
. . 33 5 Standard Deviation of DCT Coefficients for 1 < i < 6 . . . . . . . . . . . .

SUBSTITUTE SHE~'T

WO 92110830 ~ ~ ~ ~ ~ ~ ~ PCTlUS91109135 6 Division of Prediction Residuals into Blocks in Encoding Example . . . . .

r Quantizers for Q; in Encoding Example . . . . . . . . . . . . . . . . . . .
. 46 8 Quantizers for C;~ in Encoding Example . . . . . . . . . . . . . . . . . . .
. 47 9 Construction of u; in Encoding Example ( 1 of 3) . . . . . . . . . . . . . .
. 48 Construction of u; in Encoding Example (2 of 3) . . . . . . . . . . . . . . .

11 Construction of u; in Encoding Example (3 of 3) . . . . . . . . . . . . . .
. 50 12 Breakdown of Algorithmic Delay . . . . . . . . . . . . . . . . . . . . . .
. . 57 SUBSTITUTE SHEEP

WO 92/10830 '~ ~ ~ ~j ~ '~ ~ PCT/US91/09135 1 Introduction This document provides a complete functional description of the I~IIviARSAT-M
speech coding algorithm. This document describes the essential operations which are necessary and sufficient to implement this algorithm. It is recommended that implementations begin with a high-level language simulation of the algorithm, and then proceed to a real-time implementation using a floating point digital signal processor such as the AT&T DSP32C, Motorola 96002 or TI TMS320C30 (2). In addition it is highly recommended that the references be studied prior to the implementation of this algorithm.
The I:~ ViARSAT Vi speech coder is based upon the Improved ~fulti-$and Excitation (I:~iBE) speech coder (7). This coder uses a new robust speech model which is referred to as the '~iulti-Band Excitation (MBE) speech model (5). The basic methodology of the coder is to divide the speech signal into overlapping speech segments (or frames) using a window such as a Kaiser window. Each speech frame is then compared with the underlying speech model, and a set of model parameters are estimated for that particular frame. The encoder quantizes these model parameters and transmits a bit stream at 6.4 kbps. The decoder receives this bit stream, reconstructs the model parameters, and uses these model parameters to generate a synthetic speech signal. This synthesized speech signal is the output of the IMBE speech coder as shown in Figure 1.
The IMBE speech coder is a model-based speech coder, or vocoder. This means that the I:~iBE speech coder does not try to reproduce the input speech signal on a sample by sample basis. Instead the IMBE speech coder constructs a synthetic speech signal which contains the same perceptual information as the original speech signal. Many previous vocoders (such as LPC vocoders, homomorphic vocoders, and channel vocoders) have not been successful in producing high quality synthetic speech. The IMBE speech coder has two primary advantages aver these vocoders. First, the IMBE speech coder is based on the MBE speech model which is a more robust model than the traditional speech models used in previous vocoders. Second, the IMBE speech coder uses more sophisticated algorithms to estimate the speech model parameters, and to synthesize the speech signal from these model parameters.
This document is organized as follows. In Section 2 the :~iBE speech model is briefly SUBSTITUT~SNEET

WO 92/10830 !~ ~ ~ PCT/US91/09135 Transmitter ..,..,....,...,.......,...,...,..,....................,........., Analysis Quantization and Encoding Speech ; Speech Model Bit Stream Parameters : at 6.4 kbps Receiver ~~~ Synthesis ~ Decoding and Reconstruction Synthesized , I Speech Model Bit Stream Speech Parametero at 6.4 kbpa Figure 1: Improved Multi-Band Excitation Speech Coder discussed. Section 3 examines the methods used to estimate the speech model parameters, and Section 4 examines the quantization, encoding, decoding and reconstruction of the VIBE model parameters. The error correction and the format of the 6.4 kbps bit stream is discussed in Section 5. This is followed by an example in Section 6, which demonstrates the encoding of a typical set of model parameters. Section 7 discusses the synthesis of speech from the hiBE model parameters. A few additional comments on the algorithm and this document are provided in Section 8. The attached appendices provide necessary information such as the initialization for parameters. In addition Appendix J
contains Bow charts for some of the algorithms described in this document.
2 Multi-Band Excitation Speech Model Let s(n) denote a discrete speech signal obtained by sampling an analog speech signal. In order to focus attention on a short segment of speech over which the model parameters are assumed to be constant, a window tv(n) is applied to the speech signal s(n).
The windowed SUBSTITUTE SHEET

WO 92/10830 ~ ,~ ~ ~ /~ '~, C PCT/US91/09135 c Ls ~ ~ r speech signal sW(n) is defined by sw(n) - s(n)w(n) (I) The sequence s,~(n) is referred to as a speech segment or a speech frame. The IMBE analysis algorithm actually uses two different windows, wR(n) and w~(n), each of which is applied separately to the speech signal via Equation (1). This will be explained in more detail in Section 3 of this document. The speech signal s(n) is shifted in time to select any desired segment. For notational convenience s,~(n} refers to the current speech frame.
The next speech frame is obtained by shifting s(n) by 20 ms.
A speech segment s,~(n) is modelled as the response of a linear filter h",(n) to some excitation signal ew(n). Therefore, S~,(~), the Fourier Transform of s,~(n), can be expressed as Sw(~) = R,~(~)E,~(~) (2) where Hw(ca) and E,~(~) are the Fourier Transforms of h,~(n) and e,~(n), respectively.
In traditional speech models speech is divided into two classes depending upon the nature of the excitation signal. For voiced speech the excitation signal is a periodic impulse sequence, where the distance between impulses is the pitch period Po. For unvoiced speed the excitation signal is a white noise sequence. The primary differences among traditional vocoders a,re in the method in which they model the linear filter hw(n). The spectrum of this filter is generally referred to as the spectral envelope of the speech signal. In a LPC
vocoder, for example, the spectral envelope is modelled with a low order all-pole model.
Similarly, in a homomorphic vocoder, the spectral envelope is modelled with a small number of cepstral coefficients.
The primary difference between traditional speech models and the MBE speech model is the excitation signal. In conventional speech models a single voiced/unvoiced (V/UV) decision is used for each speech segment. In contrast the MBE speech model divides the exutation spectrum into a number of non-overlapping frequency bands and makes a V/UV
decision for each frequency band. This allows the exutation si&nal for a particular speech segment to be a mixture of periodic (voiced) energy and noise-like (unvoiced) energy. This added degree of freedom in the modelling of the exutation allows the MBE
speech model SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ U ''~ ? ,~] PCT/US91/09135 Voiced Speech Unvoiced Speech Speech Spectrum Spectrum Spectrum f f f Traditional Speech MBE Speech Model Model (voiced) f f Figure 2: Comparison of Traditional and MBE Speech Models to generate higher quality speech than conventional speech models. In addition it allows the i'iBE speech model to be robust to the presence of background noise.
In the MBE speech model the excitation spectrum is obtained from the pitch period (or the fundamental frequency) and the V/UV decisions. A periodic spectrum is used in the frequency bands declared voiced, while a random noise spectrum is used in the frequency bands declared unvoiced. The periodic spectrum is generated from a windowed periodic impulse train which is completely determined by the window and the pitch period. The random noise spectrum is generated from a windowed random noise sequence.
A comparison of a traditional speech model and the MBE speech model is shown in Figure 2. In this example the traditional model has classified the speech segment as voiced, and consequently the traditional speech model is comprised completely of periodic energy.
The MBE model has divided the spectrum into 10 frequency bands in this example. The fourth, fifth, ninth and tenth bands have been declared unvoiced while the remaining bands have been declared voiced. The excitation in the MBE model is comprised of periodic energy only in the frequency bands declared voiced, while the remaining bands are comprised of SUBSTfTUTE SHEET

WO 92/ 10830 ~ ~ ~ ~ ~ ~ ~ PCT/US91 /09135 noise-like energy. This example shows an important feature of the MBE speech model.
~iamely, the V/UV determination is performed such that frequency bands where the ratio of periodic energy to noise-like energy is high are declared voiced, while frequency bands where this ratio is low are declared unvoiced. The details of this procedure are discussed in Section 3.2.
SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ ~ PCf/US91/09135 w t(n) LXscrete ~~.Pass Low.P~ Initial Pltc~
F~ter Fiher dot S(pal s(n) ~ ~ pt Ptteh Rel7ncroeat wRlnJ
VdeedllJa~deed Yt Dstermtaatlon ~' lSkSK
..
M, Figure 3: I:~4BE Speech Analysis Algorithm 3 Speech Analysis This section presents the methods used to estimate the :~iBE speech model parameters. To develop a high quality vocoder it is essential that robust and accurate algorithms are used to estimate the model parameters. The approach which is presented here differs from con-ventional approaches in a fundamental way. Typically algorithms for the estimation of the excitation parameters and algorithms for the estimation of the spettral envelope parameters operate independently. These parameters are usually estimated based on some reasonable but heuristic criterion without expDcit consideration of how close the synthesised speech will be to the original speech. This can result in a synthetic spectrum quite different from the original spectrum. In the approach used in the IMBE speech coder the excitation and spec-tral envelope parameters are estimated simultaneously, so that the synthesized spectrum is closest in the least squares sense to the on~nal speech spectrum. This approach can be viewed as an "analysis-by-syn~~esis" method. The theoretical derivation and justification of this approach is presented in references (5,6,8.
A block diagram of the analysis algorithm is shown in Figure 3. The MBE speech model parameters which mast be estimated for each speech frame are the pitch period (or equiva-SUBSTITUTE SHEET

l0 .2D
o ' i _Sp y i. . .... ~... ,... .... ....
lOZ ~~10~ ~~10~ ~~10~ "10~ ~~IOs f (liz.) Figure 4: High Pass Filter Frequency Response at $ kHz. Sampling Rate lently the fnnda,mental frequency), the V/UV decisions, and the spectral amplitudes which characterize the spectral envelope. A discrete speech signal is obtained by sampling an ana-log speech signal at 8 kHz. The speech signal should be scaled such that the maximum and minimum sample values are in the ranges (16383, 32767) aad (-32768, -16385J, respectively.
In addition any non-linear companding which is introduced by the sampling system (such as a-law or n-law) should be removed prior to performing speech analysis.
The discrete speech signal is first passed through a discrete high-pass filter with the following transfer function.
_ -i A(Z) li 99z-1 (3) Figure 4 shows the frequency response of the filter specified in equation (3) assuming an 8 kHz. sampling rate. The resulting high-pass filtered signal is denoted by s(n) throughout the remainder of this settion.
The organization of this section is as follows. Section 3.1 presents the pitch estimation algorithm. The VJUV determination is discussed in Section 3.2, and Section 3.3 discusses the estimation of the spectral amplitudes.
suesrirurE s~r~~r WO 92/10830 ~ ~ ~ ~ t~ ~ ~ PCT/US91/09135 Prsvioua Framaa Future Frart»a Pn aan t Frame t-fOm~ t~l0ma. t~1 ttl0m~. tv.Itimr- t+60mr.
Plteh Vatws: P.i P.t Pa P, Pi Error Function: E_~(P) E.t(P) EdP) Et(P) E2(P) Figure 5: Relationship between Speech Frames 3.1 Pitch Estimation The objective in pitch estimation is to determine the pitch Po corresponding to the "current"
speech frame s,~(n). Po is related to the fundamental frequency wo by Po = 2x (4) wo The pitch estimation algorithm attempts to preserve some continuity of the pitch between neighboring speech frames. A pitch tracking algorithm considers the pitch from previous and future frames, when determining the pitch of the current frame. The next speech frame is obtained by shifting the speech signal a(a) "left" by 160 samples (20 ms.) prior to the application of the window in Equation ( 1 ). The pitches corresponding to the next two speech frames are denoted by Pl and P~. Similarly, the previous frame is obtained by shifting a(n) "right" by 160 samples prior to the application of the window. The pitches corresponding to the previous two speech frames are denoted by P_~ and P_z. These relationships are shown in Figure 5.
The pitch is estimated using a two-step procedure. First an initial pitch estimate, denoted by Pl, is obtained. The initial pitch estimate is restricted to be a member of the set {21, 21.5, ... 113.5, 114}. It is then refined to obtain the final estimate of the fundamental frequency coo, which has one-quarter-sample accuracy. This two-part procedure is used in part to reduce the computational complexity, and in part to improve the robustness of the pitch estimate.
One important feature of the pitch estimation algorithm is that the initial pitch estima-SUBSTITUTE SHEET

WO 92/ 10830 ~ ~ ~ i3 ~ ~ ~ PCT/US91 /09135 wt(n) -t d0 140 40 ma.
wR(n) s.(n) -tto tto Figure 6: Window Alignment lion algorithm uses a different window than the pitch refinement algorithm.
The window used for initial pitch estimation, tvl(n), is 281 samples long and is given in Appendix B. The window used for pitch refinement (and also for spectral amplitude estimation and V/UV
determination), wR(n), is 221 samples long and is given in Appendix C.
Throughout this document the window functions are assumed to be equal to zero outside the range given in the Appendices. The center point of the two windows must coincide, therefore the first non-zero point of mR(n) must begin 30 samples after the first non-zero point of wf(n).
This constraint is typically met by adopting the convention that tuR(n) = mR(-n) and u:~(n) = wl(-n), as shown in Figure 6. The amount of overlap between neighboring speech segments is a function of the window length. Specifically the overlap is equal to the window length minus the distance between frames (160 samples). Therefore the overlap when using tuR(n) is equal to 61 samples and the overlap when using wl(n) is equal to 121 samples.
SUBSTITUTE SHEET
Frarrw o Frame t o ~ Fnrtw Z
20 ma.

WO 92/10830 ~ ~ ~ ~ ~ ~ ~ PCT/US91109135 3.1.1 Determination of E(P) To obtain the initial pitch estimate an error function, E(P), is evaluated for every P in the set {21, 21.5, ... 113.5, 114). Pitch tracking is then used to compare the evaluations of E(P), and the best candidate from this set is chosen as P~. This procedure is shown in Figure 7. The function E(P) is defined by t4o sz wz _ p , l~'°'1 ,.o r(n ~ P
E ~.~=_14o LPF(J) I(J) ~n=-ill ) (5) (P) (~~40 ~40JLPF(J)wl(J)~(1 - P' ~~~ mowl(J)~
where w~(n) is normalized to meet the constraint t4o (6) w!(7) = 1.0 ~=-i4o This constraint is satisfied for wl(n) listed in Appendix B. The function r(t) is defined for integer values of t by uo r(t) _ ~ sLPF(J)wl(J)sLPF(J + t)wj(J '~ t) (7) ~=-14o The function r(t) is evaluated at non-integer values of i through linear interpolation:
r(t) =(1+~t) -t)~r(~tJ)+(t- ltJ)'r(ltJ +1) (8) where ~rJ is equal to the largest integer less than or equal to r (i.e.
truncating values of x). The low-pass filtered speech signal is given by to sLPF(n) _ ~ s(n - J)hLPF(J) ~--to where hLpF(n) is a 21 point FIR filter given in Appendix D.
The function E(P) in Equation (5) is derived in (5,8). The initial pitch estimate Pj is chosen such that E(PJ) is small, however, P~ is not chosen simply to minimize E(P).
Instead pitch tracking must be used to account for pitch continuity between neighboring speech flames.
3.1.2 Pitch lacking Pitch tracking is used to improve the pitch estimate by attempting to limit the pitch devia-tion between consecutive frames. If the pitch estimate is chosen to strictly minimize E(P), SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ ~ ~ PCT/US91/09135 Look-8vdc Pv Pitch TncldnQ
Caiculat~ ~) l:omp~rison of S(n)wt(n) --~ E~ Functlo~ ~ ( Pitch E.sHmat~s Look~Ahwd Pitch Tnddn9 Pr Figure 7: Initial Pitch Estimation then the pitch estimate may change abruptly between succeeding frames. This abrupt change in the pitch can cause degradation in the synthesized speech. In addition, pitch typically changes slowly; therefore, the pitch estimates from neighboring frames can aid in estimating the pitch of the current frames.
For each speech frame two different pitch estimates are computed. The first, PB, is a backward estimate which maintains pitch continuity with previous speech frames. The sec-ond, PF, is a forward estimate which maintains pitch continuity with future speech frames.
The backward pitch estimate is calculated with the look-back pitch tracking algorithm.
while the forward pitch estimate is calculated with the Look-ahead pitch tracking algorithm.
These two estimates are compared with a set of decision rules defined below, and either the backward or forward estimate is chosen as the initial pitch estimate, P~.
3.1.3 Look-Back Pitch 'Lacking Let P_t and P ~ denote the initial pitch estimates which are calculated during the analysis of the previous two speech frames. Let E_1(P) and E_z(P) denote the error functions of Equation (5) obtained from the analysis of these previous two frames as shown in Figure 5.
Then E_1(P_t) and E_z(P_Z) will have some specific values. Upon iaitialization the error functions E_1(P) and E_~(P) are assumed to be equal to zero, aad P_t and P_z are assumed to be equal to 100.
Since pitch continuity with previous frames is desired, the pitch for the current speech frame is considered in a raage near P_t. First, the error function E(P) is evaluated at each sussrirurE sH~~.

WO 92/10830 ~ ~~ ~ b ~ ~ j PCT/US91/09135 value of P which satisfies constraints ( 10) and ( 11 ).
.8P_t < P _< 1.2P_t (10) P E {21,21.5,...113.5, 114} (11) These values of E(P) are compared and PB is defined as the value of P which satisfies these constraints and which minimizes E(P). The backward cumulative error CEB(PB) is then computed using the following formula.
CEB(PB) = E(PB) + E_~(P_t) + E_z(P z) (12) The backward cumulative error provides a confidence measure for the backward pitch esti-mate. It is compared against the fota~ard cumulative error using a set of heuristics defined in Section 3.1.4. This comparison determines whether the forward pitch estimate or the backward pitch estimate is selected as the initial pitch estimate for the current frame.
3.1.4 Look-Ahead Pitch Tracking Look-ahead tracking attempts to preserve pitch continuity between future speech frames.
Let El(P) and EZ(P) denote the error functions of Equation (5) obtained from the two future speech frames as shown is Figure 5. Since the pitch has not been determined for these future frames, the look-ahead pitch tracking algorithm must select the pitch of these future frames. This is done in the following manner. First, Po is assumed to be fixed. Then the Pl and PZ are found which jointly minimize Et(Pt) + Ez(P~), subject to constraints (I3) through (16).
Pt E {21,21.5,...113.5,114} (13) .8-Po < Pl < 1.2~ Po (14) Pz E {21,21.5,...113.5,114} (15) .8~P1 < Pz < 1.2~Pt (16) The valves of Pl and Pl which jointly minimize Ei(Pt)+F~(P1) subject to these constraints are denoted by Pl and Pz, respectively. Once Pt and Ps have been computed the for~s~ard cumulative error function CEF(Po) is computed according to:
CEF(Po) = E(Po) + Et(Pt) + Ez(P~) (1' ) SUBSTITUTE SHEET

WO 92/10830 '~ ~ ~ ~ ~ '~ ~ PCT/US91/09135 This process is repeated for each Po in the set {21,21.5,...113.5, 114}. The corresponding values of CEF(Po) are compared and Po is defined as the value of Po in this set which results in the minimum value of CEp(Po).
Once P° has been found the integer sub-multiples of Po (i.e. ~, ~, ...~) must be considered. Every sub-multiple which is greater than or equal to 21 is computed and replaced with the closest member of the set {21, 21.5, ... 113.5, 114} (where closeness is measured with mean-square error). Sub-multiples which are less than 21 are disregarded.
The smallest of these sub-multiples is checked against constraints (18), (19) and (20).
If this sub-multiple satisfies any of these constraints then it is selected as the forward pitch estimate, Pp. Otherwise the next largest sub-multiple is checked against these constraints, and it is selected as the forward pitch estimate if it satisfies any of these constraints. This process continues until all pitch sub-multiples have been tested against these constraints.
If no pitch sub-multiple satisfies any of these constraints then PF = Po. Note that this procedure will always select the smallest sub-multiple which satisfies any of these constraints as the forward pitch estimate.
Po CEF( p ) CEF( n ) < .85 and CEF( po) <_ 1.7 ( 18) p CEF( n° ) < .4 and CEF( p ) < 3.5 ( 19) F( o) CEF( p° ) < .05 (20) n Once the forward pitch estimate and the backward pitch estimate have both been com-puled the forward cumulative error and the backward cumulative error are compared. De-pending on the result of this comparison either PF or PB will be selected as the initial pitch estimate P~. The following set of decision rules is used select the initial pitch estimate from among these two candidates.
If CEB(pg) < .48, then Pl = Pa (21) Else if CEB(PB) <_ CEF(PF), then P~ = PB (22) SUBSTITUTE SHEET

WO 92/10830 ~ ~ ;~ ;~ PCT/US91/09135 s(nMa(n)-"~ p~L S.(m) AOab) S.(m.nb) +
Calculate CalculaN + Caleutata A~(~ S.(m,a7a) .~ Ex(Wb) ER(~
Generate t0 (~
Pt c~ _ Candidaba IAlrtim~a ri.,. Wb t6384pt. Wre(m) oFr Figure 8. Pitch Refinement Else pl = pF (23) This completes the initial pitch estimation algorithm. The initial pitch estimation, Pl, is a member of the set {21, 21.5, ... 113.5, 114}, and therefore it has half sample accuracy.
3.1.5 Pitch Refinement The pitch refinement algorithm improves the resolution of the pitch estimate from one half sample to one quarter sample. Ten candidate pitches are formed from the initial pitch estimate. These are Pf - e, P~ - g, . . ., P~ + g, and PI + 8. These candidates are converted to their equivalent fundamental frequency using Equation (4). The error function ER(cao), defined in Equation (24), is evaluated for each candidate fundamental frequency ~o. The candidate fundamental frequency which results in the minimum value of ER(c.~o) is selected as the refined fundamental frequency ~o. A block diagram of this process is shown in Figure 8 .
11 "o' 's)'~roJ
ER(~o) _ ~ ~5,~(m) - S,~(m,wo)~~ (24) m =50 SUBSTITUTE SHEET

WO 92/10830 ,~. ~ ~ ~ ~ ~ PCT/US91/09135 The synthetic spectrum Sw(m,wo) is given by, Ao(wo)WR(64m) for (aa~ < m < (bo~
AOwo)~'R(L~m - ~~ + ~5J) for (a~J < m < (b1J
Sw(m,wo) _ (25) At(wo)WR( L64m - issa~lr.ao + .5J ) for (at~ < m < (bt~
where at, 6t and At are defined in equations (26) thru (28), respectively. The notation (x~
denotes the smallest integer greater than or equal to x.
ar - 2560 - .5)wo (26) 2a bt = 26(1 + ~5~0 (27) ~~b'~ ~a~i S~(m)~'R( lam - ts38, fwo + .5J
Ar(wo) _ (erl-> >e,~a., (28) ~,n=(orl ~ WR( ~64m - Z~ lwo + .5J )~Z
The function S,~(m) refers to the 256 point Discrete Fourier Transform of s(n) ~ wR(n), and WR(m) refers to the 16384 point Discrete Fourier Transform of wR(n). These relationships are expressed below. Reference (15~ should be consulted for more information on the DFT.
mo S,~(m) _ ~ s(n)wR(n)e-~~' for -128 < m < 127 (29) "=-iio mo i..~~
WR(m) _ ~ tvR(n~e-~'~ for -8192 < m <_ 8191 (30) n=-t io The notation WR(m) refers to the complex conjugate of WR(m). However, since wR(n) is a real symmetric sequence, WR(m) = WR(m).
Once the refined fundamental frequency has been selected from ~>>ong the ten candi-dates, it is ased to compute the number of harmonics in the current segment, L, according to the relationship:
~.ss~,~ +.2sJJ if ~,~ +.25J < 40 (31) 1.77~~ + .25J + 7.41J otherwise SUBSTITUTE SHE~'T

WO 92/10830 ~ ~ ~ ~ '~ ~ ~ PCT/US91/09135 Compute ~ t Com pare S~tm) Voicing Measure W>th Thresold t sksK
Compute ~ Update ~ Compute ~ -Threshold ~..~. ~~ ~ ~.a Figure 9: IMBE Voiced/Unvoiced Determination In addition the parameters a~ and b~ for 1 <_ J < L are computed from c.:ro according to equations (32) and (33), respectively.
a, = 2~6(f - .5~:ro (32) bi = ?x6 (t +.S~O (33) 3.Z Voiced/Unvoiced Determination The voiced/unvoiced (V/UV) deasions, vk for 1 < k < K, are found by dividing the spectrum into IC frequency bands and evaluating a voicing measure, Dk, for each band.
The number of frequency bawds is a function of L wad is given by:
t+s if L < 36 (34) 12 otherwise The voicing measure for 1 < k < K - 1 is given by Dk ' ~fb~r~a~~_~l (S'~(m) - S'~(m~"o)~z (35) ~f~.~a~~_~1 ~S'~(m)~z where ~ is the refined fundamental frequency, and fii, 6~, S~,(m), and S~,(m,cao) are defined in section 3.1.5. Similarly, the voicing measure for the highest frequency band is given by ~SW(m) - Sw(m~~ao)~Z
D _ ,~= fa,k_>> ( 36 ) JC
fe ~-i ~",~ fa~x_,l IS,~(~)~z suBST~ruTE s~~r WO 92/10830 '~ ~ ~ ~ ~ ~ ~ PCT/US91/09135 The parameters Dk for 1 < k < K are compared with a threshold function Of(k,c:y) given by OE(k,";;o) _ (.35 + .557tvo)(1.0 - .4775(k - l~.:ao) ~ M(~o,~avy,~mw, ~mas) (37) The parameter ~o is equal to the energy of the current segment, and it is given by ll ~'-.sJ i:ø,:oJ
~o = ~ ~5,~(m)~~ (38) m.0 The parameters ~avg, ~mas and ~m;n roughly correspond to the local average energy, the local maximum energy and the local minimum energy, respectively. These three parameters are updated each speech frame according to the rules given below. The notation ~a"9(0), ~maz(0) and ~m;n(0) refers to the value of the parameters in the current frame, while the notation ~Q"9(-1), ~.n~z(-1) and ~,nin(-1) refers to the value of the parameters in the previous frame.
- .7 ~a~y(-1) + ~3 ~o (39) ~mas(0) - .5 ~mns(-1) ~' .5 ~0 if Ep > ~mas(-1) ( 0) -99 ~ma=(-1) + .O1 ~o otherwise .5 min(-1) ~' .5 ~p ~f ~0 ~ min(-1) ~min(0) - .975~min(-1)'f' .025~p if ~min(-1) ~ ~0 < 2~min(-1) (41) 1.025 ~m;n(-1 ) otherwise After these parameters are updated, the two constraints expressed below are applied to ~min(0) and ~mas(0)~
Emin(0) - 200 if ~,n;n(0) < 200 (42) ~m~(0) - 20000 if ~,n~(0) < 20000 (43) The updated energy parameters for the current frame are used to calculate the function .tt(~o,~av9,fm;n,Fmas). For notational convenience ~avD, ~m;n, and ~m~ refer to ~avg(0), ~m~n(0) and ~mas(0), ~Pectively.
.5 if ~Q~9 < 200 Eo+f.."., s Eo+f,n.= i f ~a~D > 200 l~(~O,~av9,~min,~mar) = E°t'~Sf~.a fo+E,..a - ( ) and min < .0075 ~mnr 1.0 otherwise SUBSTITUTE SHEET

~;~9~~~~
WO 92/10830 ~ PCT/US91J09135 LS36, 3K-2SL<_3K
M ~ Mz Mj ................................ M~~ htc m ..........
n ci~ 2ci~ 3t~ ,...."........,........, (L-I)ci~ Lcc~ n ........ U
Band 1 Band 2 Band K-1 Band K
Freauency Bandy Figure 10: IMBE Frequency Band Structure The function :l~f(~o,~Q"9,~minrfmar) is used in Equation (37) to calculate the V/L'V thresh-old function. If Dk is less than the threshold function then the frequency band a3k_i < c.~ <
b3k is declared voiced; otherwise this frequency band is declared unvoiced. ~
block diagram of this procedure is shown in Figure 9. The adopted convention is that if the frequency band a3k-s < ~ < 63k is declared voiced, then vk = 1. Alternatively, if the frequency band ask-Z ~ ~ < ~k is declared unvoiced, then vk = 0.
With the exception of the highest frequency bawd, the width of each frequency band is eqnaT to 3wo. Therefore all but the highest frequency band contain three harmonics of the refined fundamental frequency. The highest frequency band (as defined by Equation (36))may contain more or leas than three harmonics of the fundamental frequency. If a particular frequency band is declared voiced, then all of the harmonics within that band are defined to be voiced harmonics. Similarly, if a particular frequency band is declared unvoiced, then all of the harmonics within that band are defined to be unvoiced harmonics.
3.3 Estimation of the Spectral Amplitudes Once the V~[;V decisions have been determined the spectral envelope can be estimated as shown in Figure 11. In the IMBE speech coder the spectral envelope in the frequency band a3k-2 ~ ~ < ~k is specified by 3 spectral amplitudes, which are denoted by M3k_~, .ll3k-i, SUBSTITUTE SHEET

WO 92/ 10830 ~ ~ ~ ~~ ~ ~ ~ PCT/US91 /09135 S.(m) Calculate Unvoiced c~u Amplitude Select Voiced or Unvoiced Spectral Amplitude 151.
S"(m) Calculate voiced ci~ Amplitude l5ks Y
Figure 11: IMBE Spectral Amplitude Estimation and .tt3~. The relationship between the frequency bands and the spectral amplitudes ie shown in Figure 10. If the frequency band a3k-z < c~ < 63k is declared voiced, then M3k-Z.
.t~t3k_r, and :'~t3k are estimated by, Mr = IAr(~:~o)I (45) for I in the range 3k -2 < l < 3k and where Ar(wo) is given in Equation (28).
Alternatively, if the frequency band 83k-Z < ~ < 63k is declared unvoiced, then M3k-s, Msk-1, and M~
are estimated according to:
__ 1 ~~~1-1 ~S~(rn)Iz ~
mc(G~1 [~nc~-llowR(n)] ~ (~61~ - larl) ( ) forlintherange3k-2<l<3k.
This procedure must be modified slightly for the highest frequency band which covers the frequency interval a3k-~ < c.r < 6L. The spectral envelope in this frequency band is represented by L - 3fC + 3 spectral amplitudes, denoted M3k-z, M3k-i ~ . .
., ML. Ii this frequency band is declared voiced then these spectral amplitudes are estimated using equation (45) !or 3K - 2 < l <- L. Alternatively, if this frequency band is declared unvoiced then these spectral amplitudes are estimated using equation (46) for ~K - 2 <
l < L.
As described above, the spectral amplitudes M1 are estimated in the range 1 <
I < L, where L is given in Equation (31). Note that the lowest frequency band, al <
ca < 63, is specified by Ml, btz, and M3. The D.C. spectral amplitude, Mo, is ignored in the IMBE
speech coder and can be assumed to be zero.
SUBSTITUTE SHEfT

Parameter ,'Vumber of Bite Fundamental Frequency8 Voiced/Unvoiced DecisionsK

Spectral Amplitudes 75 - K

Table 1: Hit Allocation Among Model Parameters 4 Parameter Encoding and Decoding The analysis of each speech frame generates a set of model parameters consisting of the fundamental frequency, ~Q, the V/UV decisions, vk for 1 < k < K, and the spectral amplitudes, .41j for 1 < l < L. Since this voice codec is designed to operate at 6.4 kbps with a 20 ms. frame length, 128 bits per frame are available for encoding the model parameters.
Of these 128 bits, 45 are reserved for error correction as is discussed in Section 5 of this document, and the remaining 83 bits are divided among the model parameters as shown in Table 1. This section describes the manner in which these bits are used to quantize, encode, decode and reconstruct the model parameters. In Section 4.1 the encoding and decoding of the fundamental frequency is discussed, while Section 4.2 discusses the encoding and decoding of the V/UV decisions. Section 4.3 discusses the quantization aad encoding of the spectral amplitudes, and Section 4.4 discusses the decoding and reconstruction of the spectral amplitudes. Reference (9J provides general information on many of the techniques used in this section.
4.I L~ndamental Frequency Encoding and Decoding The fundamental frequency is estimated with one.quarter sample resolution in the inter-°~ msus ~ ~ < 19 875 ~ however, it is only encoded at half-sample resolution. This is accomplished by finding the value of 6o which satisfies:
60 = (~ _ 39~ (47) The quantity bo can be represented with 8 bits using the unsigned binary representation shown in Table 2. This binary representation is used throughout the encoding and decoding SUBSTITUTE SHEET

~~9~!~ ~~~

valuehits Table 2: Eight Bit Binary Representation of the IV1BE model parameters.
The fundamental frequency is decoded and reconstructed at the receiver by using Equa-tion (48) to convert 6o to the received fundamental frequency c:ro. In addition bo is used to calculate K and L, the number of V/UV decisions and the number of spectral amplitudes, respectively. These relationships are given in Equations (49) and (50).
_ 4a bo + 39.5 (48) ~.96~~ + .25J J if ~~ + .25J < 40 (49) ~.77 ~~ + .25 J + 7.41 J otherwise ifL<36 K = (50) 12 otherwise Since K and L control subsequent bit allocation by the receiver, it is important that they equal K and L, respectively. This occurs if there are no uncorrectable bit errors in the six most significant bits (MSB) of bo. For this reason these six bits are well protected by the error correction scheme discussed in Section 5. A block diagram of the fundamental frequency encoding and decoding process is shown in Figure 12.
Vote that the encoder also uses equation (48) to reconstruct the fundamental frequency from 6o as shown in Figure 12. This is necessary because the encoder needs c:~o in order to compute the spectral amplitude prediction residuals via equations (53) through (54).
SUBSTITUTE SHEET

WO 92110830 ~ ~ ~ ~ ~ ~ ~ PCTlUS91/09135 _ Fundartwntal FrrVwncY
Encodlrp Fundarr~nui FnQu~neY
D~codfn9 Fundamental bo ~ Fr~qwncyr Drcodinq Compute 1' L& K
Figure 12: Fhndamental Frequency Encoding and Decoding 4.2 Voiced/Unvoiced Decision Encoding and Decoding The V/UV decisions v~, for 1 < k < K, are binary values which classify each frequency band as either voiced or unvoiced. These values are encoded using IC
bi = ~ vt 2is_k (51) ~=i The encoded value bl is represented with K bits using the binary representation shown in Table 2. At the receiver the K bits corresponding to bl are decoded into the V/UV decisions vk for 1 < k < fC. This is done with the following equation.
vk = I 2Kt kJ - 2 ~2Kb~1_k~ for 1 < k < K (52) If there are no uncorrectable bit errors in 61 and if L = L, then the transmitted V/UV
decisions, vk, will equal the received V/UV decisions 6k. Figure 13 shows a block diagram of the V/L'V decision encoding and decoding process.
4.3 Spectral Amplitudes Encoding WO 92/ 10830 ~ ~ ~ ~ ~ ~ ~ PCl"/US91 /09135 VNV Ovdafon ~t Encoding t sks K
VNV Dvclsion vt D~codnp i sksK
Figure 13: V/UV Decision Encoding and Decoding PRBA
_ + Ti Oivid~
hi,(0) btt s -~- Into OCT Gwnttzs l9sL . 6 Btoeka 2s15L.3 Hf9har Ordw Co~ffkwnta Raconatruct Hl9h~r Ordar ~ ~ PRBA
CoaffkNnts A1o(0) ~mP~rta My-1 ) t Mi(0) + Reform Pradictad Frama + from D
Sp~ctrd (~(-1) --~ Amplitudes ~ Dalar + 6 Blocks Figure 14: Encoding of the Spectral Amplitudes The spectral amplitudes M~, for 1 < I < L, are real values which must be quantized prior to encoding. This is accomplished as shown in Figure 14, by forming the prediction residuals Tr for 1 < I < L, according to Equations (53) through (54). For the purpose of this discussion Mf(0) refers to the unquantized spectral amplitudes of the current frame, Ml(-1) refers to the quantized spectral amplitudes of the previous frame, X0(0) refers to the reconstructed fundamental frequency of the current frame and ey(-1) refers to the reconstructed fundamental frequency of the previous frame. Upon initialization M~(-1) should be set equal to 1.0 for all (, and i.~(-1) _ .02x.
kr = ~°(0) , l (53) cap(-1) SugSTITUTE SHEET

~~?~~~~'S

.C=34 Bbtk 11 B_1~.2 H_lssikk~ 13_l~ l~_ .ø Blxk 66 c~,J c3~ c~J c3J ce,l T~T~i,~T,~T~Tf Length: J, = 5 J~ = 5 J, = 6 J, = 6 J3 = 6 J6 = 6 Low Frequency High Frequency Figure 15: Prediction Residual Blocks for L = 34 Tr = logz :1'lr(0) - .~5 ((1 + ~krJ - kl) logi.'1'fik~i(-1) + (kl - ~k~J) logz MIk~J+t(-1)) (54) In order to form T~ using equations (53) and (54), the following assumptions are always made:
:blo(-1) _ 1.0 (55) Mt(-1) - :1IL~_1~(-1) for l > L(-1) (56) The L prediction residuals are then divided into 6 blocks. The length of each block, denoted J; for 1 < i < 6, is adjusted such that the following constraints are satisfied.
s .1~ = L (57) ~m ~6~ < j; < J;+1 < (61 for 1 < i < 5 (58) The first or lowest frequency block is denoted by c~~ for 1 < j < Jl, and it consists of the first J1 consecutive elements of T~ (i.e. 1 < l < jt ). The second block is denoted by cz~ for 1 < j < JZ, and it consists of the next Jz consecutive elements of Tl (i.e.
J~+1 < 1 < J,+J~).
This continues through the sixth or highest frequency block, which is denoted by c6~ for I < j < js. It consists of the last J6 consecutive elements of Ti (ix. L + 1 -Je < l < L).
.fin example of this process is shown in Figure 15 for L = 34.
Each of the six blocks is transformed using a Discrete Cosine Transform (DCT), which is discussed in (9J. The length of the DCT for the i'th block is equal to J;.
The DCT
SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ ~ ~ PCT/CJS91/09135 ~=34 o.c. co.ma.~,e ., Block 1 ' R, 1, = 5 2'~ DCT : ~ C'.t oc~~'co.h°~'~sd.~b s D.c. co.ma.r,t ., Block 2 ' R2 PRBA Vector C~ 5 pt' 1 Cat Nfqh~r Otdvr ~z ° 5 DCT ~ ocT co.n~a.~,b t D.C. Co~ffkl~nt B~ock 6 C6~ 6 pt. i C6.t Niphw Order 1s = 6 DCT ~ cct co.ma."b s Figure 16: Formation of Prediction Residual Hlock Average Vector coefficients are denoted by C;,k, where 1 < i < 6 refers to the block number, and 1 < k < J;
refers to the particular coefficient within each block. The formula for the computation of these DCT coefficients is as follows:
C;,r~ _ ~. ~e:,~ cos(~(k ~(J ~)J for 1 < k < .I; (59) ~=t The DCT coefficients from each of the six blocks are then divided into two groups. The first group consists of the first DCT coefficient (i.e the D.C value) from each of the six blocks.
These coefficients are used to form a six element vector, R; for 1 < i < 6, where R; = C;,1.
The vector l~ is referred to as the Prediction Residual Block Average (PRBA) vector, and its construction is shown in Figure 16. The quantization of the PRBA vector is discussed in section 4.3.1.
The second group consists of the remaining higher order DCT coefficients.
These coef-ficients correspond to C;a, where 1 < i < 6 and 2 < j < j;. 'dote that if .7;
= 1, then there are no higher order DCT coefficients in the i'th block. The quantization of the higher order DCT coefficients is discussed in section 4.3.2.
One important feature of the spectral amplitude encoding algorithm, is that the spectral amplitude information is transmitted differentially. Specifically a prediction residual is su~asT~TUTE sHE~r WO 92/10830 ~ ~ ~ ~ ~ ~ ~ PCT/US91/09135 transmitted which measures the change in the spectral envelope between the current frame and the previous Game. In order for a differential scheme of this type to work properly, the encoder must simulate the operation of the decoder and use the reconstructed spectral amplitudes from the previous frame to predict the spectral amplitudes of the current frame.
The IblBE spectral amplitude encoder simulates the spectral amplitude decoder by setting L = L and then reconstructing the spectral amplitudes as discussed above. This is shown as the feedback path in Figure 14.
4.3.1 Encoding the PRBA Vector The PRB A vector is quantized using a three step procedure. First the vector mean mR
is computed using equation (60), and it is scalar quantized using the 6 bit non-uniform quantizer defined in Appendix E. Next the PRBA vector is vector quantized using the bit zero-mean codebook defined in Appendix F. Finally a six element quantization error vector, Q; for 1 < i < 6, is computed by adding the quaatized vector mean to the selected entry from the 10 bit zero-mean codebook and then subtracting the result from the nnqua.ntized PRBA vector R;. If L < 24, then the six elements of Q; are scalar quantized using unifotzn quantization. The step size and bit allocation for these uniform qua,ntizers is documented later is this section. If L'>_ 24, then Q; is disregarded.
The first step in quantizing the PRBA vector is to calculate the mean mR as defined in Equation (60).

mR - 6 ~ R; (60) i=1 The mean is scalar quantized by computing the mesa square error between the unquantized mean and each of the 64 qnantization levels listed in Appendix E. The 6 bit value 6z is defined as the index of the quantizer value (as shown in Appendix E) which is closest, in a mean-square error sense, to mR. It is represented using the same binary representation as is shown in Table 2.
After the mean is qaantized the PRBA vector is vector qua.ntized using the codebook listed in Appendix F. This is accomplished by computing the mean square error between the unquantized vector and each of the 1024 quantization vectors listed in Appendix F.
Note that each of these quantization vectors has zero mean. The 10 bit value b3 is defined SUBSTITUTE SHEET

WO 92110830 ~ ~ ~ ~ ~ PCT1US91/09135 as the index of the quantization vector (as shown in Appendix F) which yields the minimum mean-square error, and it is represented using the same binary representation as is shown in Table 2. Additional information on vector quantization can be found in (12).
Finally the quantized mean from Appendix E is added to each element of the selected zero-mean quantization vector from Appendix F. The resulting vector is then subtracted from R; to form the vector Q;. Appendix G is then used to find the bit allocation for the six elements of this vector. This appendix lists the number of bits allocated to the values bL_z through 6~3, which correspond to the elements QI through Qs, respectively. Note that if L > 24, then the bit allocation is equal to zero for these six elements. Each element of Q, is uniformly scalar quantized using a step size which is computed using Tables 3 and 4. Table 3 lists the standard deviation for the six elements of Q;, while Table 4 lists the step size for each uniform qua.ntizer as a function of the number of bits allocated to that quantizer and the standard deviation of the element. For example if Q~ is to be quantized using 3 bits, the step size is equal to .18 t .65 = .11T. If the number of bits allocated to a particular element is greater than zero, then it is uniformly quantized using equation (61).
Otherwise, if the number of bits allocated to a particular element is equal to zero, then that eleruent is assumed to be equal to zero and is not encoded.
b = ~Q') + 2$-t (61) O
The parameters 6, B and O is equation (61) refer to the bit encoding, the number of bits and the step size which has been computed for Q;, respectively. Note that if the value of b is outside the range 0 < 6 < 2~ - 1, then it is set equal to the closest value within this range. Finally, each encoding is converted into the appropriate binary representation using the same representation as is shown in Table 2.
4.3.2 Encoding the Higher Order DCT Coef$cienta Once the P RBA vector has been quantized, the remaining bits are used to encode the L - 6 higher order DCT coefficients which complete the representation of the spectral amplitudes.
Appendix G shows the bit allocation as a fraction of L for these coefficients.
For each value of L the first L - 6 entries, labeled 64 through 6L-3, provide the bit allocation for the higher SUBST1TU'TE SHEET

WO 92/10830 ~ ~ ~ ~ /~ ~ ~ PCT/US91/09135 Elemento ' 1 .25 2 .18 3 .15 4 .15 .12 6 .12 Table 3: Standard Deviation of PRBA Quantization Errors .'umber Step of Bits Siae 1 1.20 2 .850 3 .650 4 .420 5 .280 6 .140 7 .070 8 .0350 9 .01750 .008750 Table 4: Step Size of t;niform Quantizers order DCT coefficients. The adopted convention is that (b,, bs, ..., b~_3) correspond to (C1,2 r C1,3 r ~ ~ ., Cl )~ , . .., Cg,y, C6,3~ . . ., Cd J~ ~, respectively.
Once the bit allocation for the higher order DCT coefficients has been obtained, these coefficients are quantized using uniform quantization. The step size used to quantize each coefficient is a function of the bit allocation and the standard deviation of the DCT coeffi-cient. This relationship is summarized m Tables 4 and 5. For example, if 4 bits are allocated for a particular coefficient, then from Table 4 the step site, :J, equals .420. If this was the the third DCT coefficient from any block (i.e. C;,3), then 0 = .216 as shown in Table 5.
This equates to a step size of .0902. If the number bits allocated to a particular coefficient is greater than zero, then that coefficient is encoded using equation (62).
Otherwise, if the suBST~ruT~ sHE~r WO 92/10830 ~~ ~ (~ ~ ~' ~ PCT/US91/09135 DCT Coe~ciento C;,z .297 C;,a .216 .177 C;,s .165 C;,B .167 .153 C;,B .145 C;,9 .130 C;ao .130 Table 5: Standard Deviation of DCT Coefficients for 1 < i < 6 number of bits allocated to a particular coefficient is equal to zero, then that element is assumed to be equal to zero and is not encoded.
b = lC,.k'k J + 2B'' (62) D
The parameters b, B and ~ in equation (62) refer to the bit encoding, the number of bits and the step size which has been computed for C;,k, respectively. Note that if the value of b is outside the range 0 < 6 < 2~ - 1; then it is set equal to the closest value within this range. Finally, each encoding is converted into the appropriate binary representation using the same representation as is shown in Table 2.
4.4 Spectral Amplitudes Decoding The spectral amplitudes are decoded and reconstructed by inverting the quantization and encoding procedure. First six blocks are generated. The length of each block, J; for 1 <
i < 6, is adjusted to meet the following constraints.

J; = L (63) ;m ~6J<!;<J;+~<(s~ forl<i<5 (64) The elements of these blocks are denoted by C;,k, where 1 < i < 6 denotes the block number and where 1 < k < J; denotes the element within that block. The first element of each SUBSTITUTE SHEET

WO 92/ 10830 ~ ~ 9 b ~ ~ PCT/ C)S91 /09135 block is then set equal to the decoded PRBA vector R, via equation (65). The formation of the decoded PRBA vector is discussed in Section 4,4.1.
C;,i - R; for 1 < i < 6 (65) The remaining elements of each block correspond to the decoded higher order DCT coefft-dents which are discussed in Section 4.4.2.
Once the DCT coefficients C;,,t have been reconstructed, an inverse DCT is computed on each of the six blocks to form the vectors c;~. This is done using the following equations furl<i<6.
~;J = ~ a(k)C;.k cos(~(k ~(~ ! ~)) for 1 < j < j; (66) k=1 1 ifk=1 a(k) _ (67) 2 otherwise The six transformed blocks c;,~ are then joined to form a single vector of length L, which is denoted Tt for 1 < I < L. The vector Tt corresponds to the reconstructed spectral amplitude prediction residuals. The adopted convention is that the first Jl elements of Tt are equal to cl~ for 1 < j < Jl. The next Jz elements of Tt are equal to c~~ for 1 < j <
Js. This continues antil _the last Js elements of Tt are equal to cs,~ for 1 < j < Ja.
Finally, the reconstructed spectral amplitudes for the current frame are computed using the following equations.
k _ ~:ao(U) .l 68 t ~(-1) ( ) Mt(0) = exp~ln2~(Tt+.75((1+~k,J-kt)logzMlk,)(-1)+(kt-~kt~)l~zMik~J+t(-1)))I
(69) L'pon initialization !tlt(-1) should be set equal to 1.0 for all I, and c:ro(-1) _ .02x. In order to reconstruct .~fi(0) using equations (68) and (69), the following assumptions are always made:
:vto(-1) - 1.0 (70) :'l~li(-1) - ML~_~l(-1) for I > L(-1) (71) The last step is that the spectral amplitudes are enhanced using the techniques discussed in Section 4.4.3. A block diagram of the spectral amplitude decoder is shown in Figure 17.
SUBSTtTUTS SHEET

PRBA
Reform + Spectral 5i Reconstruct DC1'' from 2x Arnplftuds 2SisL;3 6 ~«~ + U ~~«~t 15ISL
HIqMr Order Coofficl~b UMnhan~d M,(0) 1s15L
luo(0) Pndicbd t Fnrt»
_ f~ :
Ct~(-1) -~ Arnp itudu Figure 17: Decoding of the Spectral Amplitudes In order for the II~fBE speed coder to operate properly the encoder and decoder must each be using the same bit allocation and step sizes. As is discussed above the encoder varies the bit allocation and step sizes depending on L, the number of spectral amplitudes in a particular frame. The encoder uses L to $nd the bit allocation from Appendix G, while the decoder uses L to find the bit allocation from Appendix G. Consequently the property that L = L is extremely critical for the proper operation of the voice codec.
The voice coding algorithm has been designed such that this property is always maintained, except in the presence of a very large number of bit errors.
4.4.1 Decoding the PRBA Vector The PRBA is decoded by using the 6 bit value bz as an index into the quantizer values listed in Appendix E. The vector mean mR is decoded as the quantizer value corresponding to this index. Similarly, the 10 bit value b3 is interpreted as an index into the quantization vectors listed in Appendix F. The decoded vector mean mR is added to each element of the quantization vector corresponding to this index, and the result is then added to the decoded quantization error vector to form the decoded PRBA vector which is denoted R,.
The quantization error vector Q; is decoded by first using L and Appendix G to find the bit allocation for b~_Z through bL+s, which corresponds to the elements Q1 and Qs, respectively.
Note that if L > 24, then the bit allocation is equal to zero for these six elements. Next the SUBSTITUTE SHEET

qnantizer step size is computed for each element of Q; using Tables 3 and 4.
This is done in the same manner as is discussed in Section 4.3.1. If the number of bits allocated to a particular element is greater than zero, then that element is decoded using equation (72).
Otherwise, if the number of bits allocated to a particular element is equal to zero, then that element is set equal to zero.
Q; = p (b _ 2B-i + .5) (72) The parameters b, B and D in equation (72) refer to the bit encoding, the number of bits and the step size which has been computed for Q;, respectively.
4.4.2 Decoding the Higher Order DCT Coefficients The higher order DCT coefhdents, which are denoted by C;,k for 2 < i < 6 and 1 < k < J;, are reconstructed from the bit encodings b4, 65, ..., 6L-3. First the bit allocation table listed in Appendix G is used to determine the appropriate bit allocation. The adopted convention is that (b,,, 6s, ..., b~_3J correspond to (CI,Z, Cl,~, ..., Ct,~y, ..., Cs,Z, Cs,~, ..., Ce,~s), respectively. Once the bit allocation has been determined the step sizes for each C;,k are computed using Tables 4 and 5. If the number of bits allocated to a particular coefficient is greater than zero, then that element is decoded using equation (73).
Otherwise, if the number of bits allocated to a particular coefficient is equal to zero, then that element is set equal to zero.
Ca.k = ~ (b - ~-l + .5) (73) The parameters b, B and O in equation (73) refer to the bit encoding the number of bits and the step size which has been computed for C;,k, respectively.
4.4.3 Spectral Amplitude Enhancement The I1SBE speech decoder attempts to improve the perceived quality of the synthesized speech by enhancing the spectral amplitudes. The unenhanced spectral amplitudes are required by future frames in the computation of Equation (69). However, the enhanced spectral amplitudes are used in speech synthesis. The spectral amplitude enhancement is SUBSTITUTE SHEET

WO 92/ 10830 ~ ~ PCT/US9l /09135 accomplished by generating a set of spectral weights from the model parameters of the current frame. First Ro and R1 are calculated as shown below I, Ro = ~ Mi ( 74 ) r=i G
R~ _ ~ Mj cos(c:~ l ) (75) m Next, the parameters Ro, and R1 are used to calculate a set of weights, Wi, given by i _ . ~.96~r(Ro + Ri - 2RaR1 cos(c;ro 1))1 fort<I<L (76) W ~ M~ ~o Ro ( Ro - R i These weights are then used to enhance the spectral amplitudes for the current frame according to the relationship 1.2~M~ if W~> 1.2 ,t~~ = fo r 1 < I < L
W~ ~ M~ otherwise (77) For notational simplicity the weighted and unweighted spectral amplitudes are both referred to as Mi in Equation (77). As mentioned previously, the unenhanced spectral amplitudes are used in the decoding the spectral amplitudes of future frames, while the enhanced spectral amplitudes are used in speech synthesis. This is shown in Figure 17.
The value of Ro corresponds to the energy of the spectral amplitudes. This value is used to update the current frame parameter Sg(0) from the previous Game parameter Sg(-1) according to the following equation.
SE(0) - .95 Sg(-1) + .051 if .95 Sg(-1) + .05 Ra > 10000.0 (7g) 10000.0 otherwise The value of SE is required in Section 5.4.
SUBSTITUTE SHEET

bo, .... ~.3 Rearrange "°' " ' u' Add Error c° "" c' lnterieave Fame Bit Vectors ~ Correction ~ ~ Bid Figure 18: Error Correction Encoding Forward Error Correction and Bit Interleaving The MBE speech coder uses a number of different error protection measures in order to increase its robustness to channel degradations. The encoded model parameters, 60, bl, ... , b~~,3, are first rearranged into a new set of vectors denoted ua, ul, ..., ur.
These vectors are then protected with error correction codes to produce a set of code vectors denoted c4, c1, . . . , c; . Intra-frame bit interleaving is then used on the error correction code vectors in order to spread the effect of short burst errors. A block diagram of the error correction encoding is shown if Figure 18. The effect of bit errors is further reduced at the decoder through the use of frame repeats and adaptive smoothing of the decoded model parameters.
5.1 Error Correction Encoding The IMHE speech coder uses 83 bits per frame to encode the model parameters and 45 bits per frame for forward error correction. The 45 error correction bits are divided between parity checJcs, 23/ 12 Golay codes and 15/ 11 Hamming codes. This division is performed according to the each bits relative sensitivity to bit errors. The encoding sad decoding algorithms for the Golay a.nd Ramming codes are discussed in the open literature (10,11).
In order to ensure sufficient performance, all error correction codes should be decoded np to their maximum error correction capability. Optionally soft-decision decoding ca.n be use used to further improved performance. The 23/12 Golay code is defined by the following matrix operation, where all operations are modulo 2. The vectors c; and u; are assumed to be row vectors, where the "left" most bit is the !vISB.
$u$sT~TUTE SH~Er PCT/LJS9l /09135 c;=u;~

Similarly, the 15/11 Hamming code is defined in following matrix operation, where all operations are modulo 2. The vectors c; and u; are assumed to be row vectors.

c;=u;~ 0 1 0 1 0 1 0 In both of the preceding matrices, absent entries ase assumed to equal zero.
gSTtTUTE SHEET
su WO 92110830 ~ ~ ~ ~ ~ ~ ~ PCT/US91/09135 l~ ~ Q~ 23112 t" °~UU
Gola Encode bo bz Q~mY
Figure 19: Format of ua and ea V I 23It 2 ~ 1 Gola Encode b~ b~, bs. ...
Figure 20: Format of ut and cl The nine most sensitive source bits are first parity checked with three bits, and all twelve bits are then encoded in a 23/12 Golay code. The nine most sensitive bits consist of the the six most significant bits of bo, and the three most significant bits of bZ.
One parity check bit is used for the three most significant bits of bo, one parity check bit is used for the next three most significant bits of 60, and one parity check bit is used for the three most significant bits of 6~. A parity check bit is computed as the modulo 2 sum of the appropriate bits. The resulting 12 bits, denoted uo, are encoded with a Golay code into the 23 bit code word co as shown in Figure 19.
The 12 next most significant source bits, ut, are encoded with a Golay code into the 23 bit code word cl. The vector ut consists of the three most significant of b3 and the most significant nine bits from the set {b~, 65, .. ., b~_3}, where only the three most significant bits from each element are considered. For example, if each element of the set was encoded with four bits, then the three most significant bits from b~, 6s, and bs, in that order, comprise the last nine bits of ut. The format of ut and ct are shown in Figure 20.
The remaining 62 bits are divided into five I1 bit vectors, u~, ..., ie~, and one 7 bit vector, u~. The 11 bit vectors are encoded with a 15/11 Hamming code into the IS bit codewords, c~, ..., ce. The seven bit vector is not error protected, therefore cr = ur.
The vectors uz, ..., us are formed in the following manner. First a large 62 bit vector is formed by concatenating (in order) the K bit vector 61, the fourth and fifth most significant bits from 6Z, the seven least significant bits from b~, the bits from the set {b~, 65, . . . , bL+s}
SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ PCT/US9l/09135 (except the nine bits which were included in u~), the two least significant bits from bo and the least significant bit from bz. For the purpose of this discussion the most significant bit of bl is referred to as the most significant bit of this large 62 bit vector.
The vectors uZ, . . . , u~ are formed from this large 62 bit vector in the following manner. The 11 most significant bits are set equal to uz. The next 11 most significant bits are set equal to u3. This process continues until the 7 least significant bits are set equal to ul.
5.2 Bit Interleaving Intra-frame bit interleaving is used to spread short burst errors among several code words.
The minimum separation between any two bits of the same error correction code is 5 bits.
The exact order of the 128 bits in each frame is given in Appendix H. The notation in this appendix is that the bit number corresponds to the significance of a particular bit. For example, bit number 2 of ct, refers to the second to least significant bit of that code word.
The notation Qo sad to refers to the first bits to be modulated onto the Q
channel and I
channel, respectively, and the notation Q~ and h refers to the last bits to be modulated onto the Q channel and I channel, respectively.
5.3 Error Correction Decoding The error correction decoding of the received bits into the bit vectors bo, bt, . . ., bL+s 'a a straightforward inverse of the error correction encoding procedure. First Appendix H is used to map the received bits into the code vectors ca, ct, ..., c~. Then the 23 bits code words co and cl are each passed through a Golay decoder, which can correct up to three errors. The output of the Golay decoder is the 12 bit vectors uo sad icl. The panty of rio is checked to ensure that it agrees with the three parity bits which were added to the input. If the panty does not check then a frame repeat is performed, as described in Section 5.4. If the parity does chec3c, then the six most significant bits of 6o and the three most significant bits of bZ are recovered. Similarly the 12 bit vector ul is used to recover the three most significant bits of b3 and the nine most significant bits from the set {b~, bs, . . . , bL_3}. Next, the 15 bit code words c~ through cs are passed through a Hamming decoder, which can correct a single bit error, and the outputs are set equal to us through ua, respectively. In SUgST~TUTE S~~~

addition the seven bit vector c~ is set equal to u~. From these vectors the 62 remaining bits are recovered in accordance with the procedure described in Section 5.1.
Since bo is represented with 8 bits, it is constrained to the range 0 < bo <
255. However because of the limitation imposed by the pitch estimation algorithm, not all of the possible values of 6o represent valid pitch estimates. In particular the values 192 <
bo < 255 do not represent valid pitch estimates. If the decoder receives a value of 6o in this range, it should not continue decoding the model parameters, but should instead take some alternative action. Specifically, if the decoder receives a value of bo in the range 192 <
bo < 199 or 204 < ba < 255 a frame repeat should be performed as is described in Section 5.4.
Alternatively, a received value of 6o in the range 200 < bo < 203 signifies that a silence frame should be synthesized, and the decoder should perform the appropriate muting or noise insertion.
5.4 Adaptive Smoothing The IMBE speech decoder estimates the number of errors in each frame of data.
This estimate is obtained from the Golay and Hamming codes which are used for forward error correction. This estimate is used to adaptively smooth the V/UV decisions, and to control a frame repeat mechanism.
The decoder calculates the total number of errors which are corrected by the Golay decoders a,nd the Hamming decoders for each of the received codewords co, ct, ..., ce. The number of errors in each codeword is estimated by the number of corrected errors in that codeword. The only exception is that if any of the parity checks in uo are incorrect, then the estimate of the number of errors in co is changed from s to 7-z (for example 2 becomes 5). The estimate of the total number of errors, eT, is equal to the sum of the error estimates from each individual codeword. This estimate is then used to update the current estimate of the error rate ER(0) from the previous estimate of the error rate eR(-1), according to:
eR(0) _ .95cR(-1)+.00042eT (79) The current estimate of the error rate is used by the decoder to adaptively smooth the V/UV decisions. First an adaptive threshold MT is calculated using equation (80). Then if SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ ~ ~ ~ PCT/US9l/09135 eR(0) > .003, each decoded spectral amplitude :'~t~ for 1 < l < L is compared against .'l~T, and any spectral amplitude which is greater than J~tT is declared voiced, regardless of the decoded V/UV decisions. If ER(0) < .003 or if a spectral amplitude is less than MT, then the received V/UV decision for that spectral amplitude is left unchanged.
X5.255 $ 0 3~5 MT - ~ m3.~s~R o if eR(0) < .02 (80) 1.414 (SE(0))~315 otherwise SE(0) is defined in Equation (78) in Section 4.4.3.
A frame repeat is performed by the Ih4BE speech decoder if too many errors are found in the current frame, or if the value of 6o is outside the allowable range, or if an error is detected in the most sensitive bits. Specifically a frame repeat is performed if eR(0) < .02 and ET > 9, or if eR(0) > .1, or if eT > 12, or if 192 < 60 < 199, or if 204 <
bo < 255, or if any of the parity checks in uo are incorrect.
If a frame repeat is performed, the I'.vIBE model parameters for the current frame are replaced by the model parameters for the previous frame. Specifically, the following replacements are made:
X0(0) _ pro(-1) (gl) L(0) - L(-1) (82) vk(0) - vk(-1) (83) M,(0) - M~(-1) (84) If a frame repeat is performed, the information contained in the bit vectors bo, bt, . . . is not used by the decoder, and speech synthesis proceeds using the replaced model parameters.
In some bad burst error conditions it is possible that several consecutive frame repeats will be required. Ia this case the model parameters received for the current frame are replaced with the model parameters from the last acceptable frame.
SUBSTITUTE SHEET

WO 92/10830 ' ~'~ ~ PCT/US91/09135 - 6? -i J; e;,l .
. .

1 2 Tl , Tz 2 2 T'3 , T, 3 3 T5, Ts, Tr 4 3 T8, T9, Tlo 3 T'11, T1 Z, 6 3 T14, T15, Tle Table 6: Division of Prediction Residuals into Blocks in Encoding Example 6 Parameter Encoding Example This section provides an example of the encoding and error correction for a typical parameter frame. In this example the fundamental frequency is assumed to be equal to: ~o - Z"
- 35.125' Since the values of L and If ate related to c:~o through equations (31) and (34), they aae equal to: L = 16 and K = 6. The remaining model parameters are left unspecified since they do not affect the numbers presented in this example.
The encoding of this example parameter frame proceeds as follows. First the fun-damental frequency is encoded into the 8 bit value bo using equation (47), and the 6 voiced/unvoiced decisions are encoded into the 6 bit value 61 using equation (51). The 16 spectral amplitude prediction residuals, T~ for 1 < I < 16, are then formed using equa-lions (53) through (54). Vext, these prediction residuals are divided into six blocks where the lengths of each block, J; for 1 < i < 6, are shown in Tahle 6. The spectral amplitude prediction residuals a,re then divided into the six vectors ci, j for 1 <_ i <
6 aad 1 < j < J;.
The first .11 elements of Tt form elf. The next .)Z elements of T~ for cs~, and so on. This is shown in Table 6. Each bIodc c;~ for 1 <_ i < 6, is transformed with a J;
point DCT
using equation (59) to produce the set DCT coefficients C;,k for 1 < k < J;.
The first DCT coefficient from each of the six blocks is used to form the PRBA vector R;, and it is qua.ntized and encoded into the 6 bit value bs and the 10 bit value 63 using appendices E
and F, respectively. The quantization error vector Q; is found by adding the quantized mean to each element of the selected codebook vector and then subtracting the result from SUBSTITUTE SHEET

WO 92/10830 PCT/L:S91/09135 ElementBif EncodingBitsStep Size Qt bt~ 2 .2125 bis 3 .1170 Q3 bls 2 .1275 Q4 b17 1 .1800 Qs bta 0 N/A

Qs bts 0 N/A

Table 7: Quantizers for Q; in Encoding Example R;. Appendix G is used to find the bit allocation for the elements of the vector Q, and this is shown in Table 7. The first four elements are uniformly quantized using equation (61), and the step sizes for these four elements are calculated from Tables 3 and 4.
The resulting step sizes are shown in Table 7. The last two elements are not qua.ntized since the number of bits allocated to those elements is equal to zero.
After the PRBA vector has been quantized and encoded, the remaining bits are dis-tributed among the ten higher order DCT coefficients, C;,k for 1 < i < 6 and 2 < k < J,.
This is done using Appendix G and the resulting bit allocation is shown in Table 8. Each DCT coefficient is then quantized using equation (61). The step sizes for these quantizers are computed using Tables 4 and 5, and the results are also shown in Table 8.
The 20 bit encodings bo through bt9 are then rearranged into the eight vectors ua through u~. This is accomplished using the procedure described in Section 5, and the result is shown in Tables 9 through 11. The convention in these tables is that the appropriate bit from the vector listed in the first two columns is set equal to the appropriate bit from the bit encoding listed in the last two columns, where the least significant bit corresponds to bit 0.
Bit 2 of vector uo is set equal to parity check of bits 11 - 9. Similarly, bits 1 and 0 are set equal to the parity eheck of bits 8 - 6 and bits 5 - 3, respectively.
The vectors uo and ut are each encoded with a 23/12 Golay code into the code vectors c-0 and ci, respectively. The five vectors uz through u~ are encoded with a 15/11 hamming code into the code vectors eZ through c~, respectively. The vector c7 is set equal to u~.
These code vectors are then interleaved as specified in Appendix H, a.nd finally the frame SUBSTITUTE SHEET

WO 92110830 ~ ~ ~ ~ ~ ~ '~ PCT~US91J09135 DCT Coe,~cientBit EncodingBilBStep Size C,,s b, 7 .02079 Cz,s 65 7 .02079 C3,z b8 5 .08316 Cs.3 ~r 5 .06048 be 4 .12474 t:,,3 b9 4 .09072 Cs,s blo 4 .12474 Cs.~ b 11 3 .14040 C6,z bli 3 .19305 b~3 3 .14040 Table 8: Quantiaers for C;~ in Encoding Example bits are transmitted to the decoder in ascending order (i.e bit 0 first and bit 127 last).
~u~s-r~TUT~ sH~~r WO 92110830 ~ ~~ ~ ~ PCT1US91/09135 VectorBit NumberhectorBit .''umber up 11 by 7 iep 10 by 6 up 9 6p 5 ib 8 6p 4 up 7 6p 3 icp 6 by 2 up 5 bZ 5 up 4 bZ 4 up 3 bz 3 up 2 parityN/A

up 1 parity,1/A

up 0 parityN/A

ul 11 b3 9 a 1 10 63 8 a l 9 b3 7 ui 8 6~ 6 ul i b~ 5 a 1 6 b~ 4 ul 5 bs 6 ut 4 bs 5 ul 3 bs 4 a 1 2 be 4 ul 1 be 3 a 1 0 6~ 2 u~ 10 6t 5 u~ 9 61 4 uz 8 bi 3 uz 7 6t 2 u~ 6 61 1 uz 5 bl 0 us 4 bz 2 uZ 3 bZ 1 uZ 2 ~ 6 u~ 1 b3 5 us 0 63 4 Table 9: Construction of u; in Encodiag Example (1 of 3) SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~~ --~~ r' ~ PCT/I,TS91/09135 VectorBit ~1'umberVectorBit .''umber us 10 b3 3 u3 9 b3 2 u3 8 bs 1 u3 7 bs 0 6 6~ 3 u3 5 6, 2 ua 4 b~ 1 us 3 b, 0 u3 2 bs 3 us 1 bs 2 u3 0 6s 1 6s 0 u~ 9 bs 1 u~ 8 b~ 0 u~ ~ ~ 4 u~ ti b7 3 u~ S b~r 2 b'r 1 b~r 0 2 bs 3 u~ 1 6a 2 u4 0 ba 1 us 10 be 0 us 9 bg 3 us 8 b9 2 us 7 be 1 a ti 69 0 s us 5 bto 3 us 4 bto 2 us 3 6to 1 us 2 bto 0 us 1 btt 2 us 0 6tt 1 Table 10: Construction of. u; in Encoding Example (2 of 3) SUBSTITUTE SHEET

VectorBit ,NumberVectorBit Number ice 10 611 0 us 9 61~ 2 us $ ~ 1 z ics 7 61s 0 us 6 613 2 ug 5 b13 1 us 4 6~3 0 us 3 61, 1 ue 2 61, 0 ice 1 615 2 ug 0 615 1 u~ 6 ~ bls 0 u~ 5 bas 1 u~ 4 bls 0 u~ 3 blt 0 ur 2 bo 1 u~ 1 by 0 u~ 0 b~ 0 Table 11: Construction of u; in Encoding Example (3 of 3) uBS,~'-ru.~~ sH~~r s WO 92/10830 ~ ~ ~ ~ ~ ~ PCTtUS91t09135 ? Speech Synthesis As was discussed in Section 3, the IMBE speech coder estimates a set of model parameters for each speech frame. These parameters consist of the fundamental frequency c:~, the V jUV
decisions for each frequency band uk, and the spectral amplitudes Mr. After the transmitted bits are received and decoded a reconstructed set of model parameters is available for synthesizing speech. These reconstructed model parameters are denoted ~, vk and Mr, and they correspond to the reconstructed fundamental frequency, V jUV decisions and spectral amplitudes, respectively. In addition the parameter L, defined as the number of spectral amplitudes in the current frame, is generated from ~a according to Equation (49). Because of quantization noise and channel errors the reconstructed model parameters are not the same as the estimated model parameters wo, vk and .'lf~.
?.1 Speech Synthesis Notation The IViBE speech synthesis algorithm uses the reconstructed model parameters to generate a speech signal which is perceptually similar to the original speech signal.
For ea.cb new set of model parameters, the synthesis algorithm generates a 20 ms. frame of speech, s(n), which is interpolated between the previous set of model parameters and the newest or current set of model parameters. The notation L(0), c:y(0), vk(0) and Mr(0) is used to denote the current set of reconstructed model parameters, while L(-1), ~(-1), 'v,~(-1) and .'I~l(-1) is used to denote the previous set of reconstructed model parameters. For each new set of model parameters, 3(a) is generated in the range 0 < n < N, where N equals 160 samples (20 ms.). This synthetic speech signal is the output of the IMBE voice coder and is suitable !or digital to analog conversion with a sixteen bit converter.
The synthetic speech signal is divided into a voiced component 3"(n) a.nd an unvoiced component a""(n). These two components are synthesized separately, as shown in Figure 21, and then summed to form 3(n). The unvoiced speech synthesis algorithm is discussed in Section i .2 and the voiced speech synthesis algorithm is discussed in Section 7.3.
For the purpose of speech synthesis each spectral amplitude is declared to be either voiced or unvoiced. The l'th spectral amplitude is declared voiced if the frequency w = ! ~~
is W thin a voiced frequency band. The frequency band (3k - 2.5);ao < ;v c (3k + .5)~o for SUBSTITUTE SHEET

Unvoiced Suv(~) (rJp LSpeech Synthesis MBE Model ~'k s(n) Parameters 1~
Synthetic Speech 1 Voiced Speech S nthesis ~' Figure 21: IVfBE Speech Synthesis 1 < k < K - 1 is voiced if 6k = l, otherwise it is unvoiced. Similarly, the highest frequency band (3K - 2.5)wo < ~ < (L + .5)r,~o is voiced if vK = 1, otherwise it is unvoiced. The I'th spectral amplitude is also declared voiced if eR > .003 and Mr > MT as is discussed in section 5.4. In all other cases the I'th spectral amplitude is declared unvoiced.
?.2 Unvoiced Speech Synthesis The energy from unvoiced spectral amplitudes is synthesized with an unvoiced speech syn-thesis algorithm. First a white noise sequence, u(n), is generated. This noise sequence can have an arbitrary mean. A recommended noise sequence (14) can be generated as shown below .
u(n + 1) = 171u(n) + 11213 - 53125~171u(a) + 112131 (85) The noise sequence is initialized to u(-105) = 3147.
For each successive synthesis frame, u(n) is shifted by 20 ms. (160 samples) and win-dowed with ms(n), which is given in Appendix I. Since ws(n) has a non-zero length of 209 samples, there is a 49 sample overlap between the noise signal used in successive synthesis frames. Once the noise sequence has been shifted and windowed, the 256 point Discrete Fourier Transform U,~(m) is computed according to:
a n tv n e'~ 1-~ for -128 < m < 127 U,~(m) _ ~ ( ) s( ) - - (86) n-_-1o4 The function U~,(m) is generated in a manner which is analogous to S~,(m) defined in SUBSTITUTE SHEET

Equation (29) except that u(n) and ws(n) are used in place of a(n) and u;R(n).
The function U,~(m) is then modified to create U,~(m). For each J in the range 1 <
1 < L(0), U~,(m) is computed according to Equation (87) if the 1'th spectral amplitude is voiced, and according to Equation (88) if the 1'th spectral amplitude is unvoiced.
U,~(m) - 0 for (ni~ < ~m~ < (b~~ ($7) U~,(m) - 1'w 3f~(0)UW(m) 1 for (aid < ~m~ < (by ($8) ~~.'«,i ~uVla)I' i tfs~l-fs~1) The unvoiced scaling coefficient 7,~ is a function of the synthesis window sus(n) and the pitch refinement window u,R(n). It is computed according to the formula:
110 ~R(n) ~ ~nw-104 ~S(n) 3 $9) 110 ~ ( n=-110 ~n=-110 u'R(~) The frequency bands edges dr and bt are computed from ~ according to equations (90) and (91), respectively.
n~ = 2~ (I - .5) ~ ~o (90) bt= 2s6(t+.5).~.'o (91) Finally, the very low frequency and very high frequenty components of U~,(m) are set equal to zero as shown in the following equation.
_, i~(m) - 0 for ~m~ < (al~ (92) 0 for (bt~ < ~m~ < 128 The seqaence u,~(n), defined as the 256 point Inverse Discrete Fourier Transform of U,~(m), is the unvoiced speech for the current frame. The sequence u~,(n) is computed as shown in the following equation.
1?7 u~,(n) _ ~6 ~ G'~,(m)e~ i~~ for -128 < n < 127 (93) =-tea - _ In order to generate ~,~"(n), uw(n) must be combined with the unvoiced speech from the previous frame. This is accomplished using the Weighted Overlap Add algorithm described SUBSTITU~'E SHEET

WO 92/10830 ~ PCT1US91109135 in (6J. If u~,(n,0) is used to denote the unvoiced speech for the current frame, and u,~(n, -1) is used to denote the unvoiced speech for the previous frame, then a""(n) is given by 'uV(n) = tvs(n)u",(n, ~1) + ws~n - N)u,~(n - N,0) (9~) u,s(n) + ms(n - N) for 0 < n < N
In this equation ws(n) is assumed to be zero outside the range -105 < n < 105, and u",(n,0) and ii",(n,-1) are assumed to be zero outside the range -128 < n <
127.
7.3 Voiced Speech Synthesis The voiced speech is synthesized by summing a voiced signal for each spectral amplitude according to the following equation.
max)L(-t ).L(o)) s"(n) _ ~ 2 ~ s~,~(n) for 0 < n < N (95) t-_ t The reader is referred to (1,5y for background information on the algorithm described in this section. The voiced synthesis algorithm attempts to match the I'th spectral amplitude of the current frame with the I'th spectral amplitude of the previous frame.
The algorithm assumes that that all spectral amplitudes outside the allowed range are equal to zero as shown in Equations (96) and (97).
Mi(0) = 0 for I > L(0) (96) Mi(-1) = 0 for l > L(-1) (97) In addition it assumes that these spectral amplitudes are unvoiced. These assumptions are needed for the case where the number of spectral amplitudes in the current frame is not equal to the number of spectral amplitudes in the previous frame (i.e. L(0) ø' L(-1)).
The signal a",~(n) is computed differently for each spectral amplitude. 1f the I'th spectral amplitude is unvoited for both the previous and current speech frame then a"a(n) is set equal to zero as shown in the following equation. In this case the energy in this region of the spectrum is completely synthesized by the unvoiced synthesis algorithm described in the previous section.
a",l(n) = 0 for 0 < n < N (98) SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ 6 4 2 ~ PCT/US91/09135 _ ?? _ Alternatively, if the !'th spectral amplitude is unvoiced for the current frame and voiced for the previous frame then 3"a(n) is given by the foi'awing equation. In this cane the energy in this region of the spectrum transitions from the voiced synthesis algorithm to the unvoiced synthesis algorithm.
s",t(n) - u~s(n) fir(-1) cos(c:~o(-1)n! + ~t(-1)~ for 0 < n < N (99) Similarly, if the t'th spectral amplitude is voiced for the current frame and unvoiced for the previous frame then ~",t(n) is wen by the following equation. In this case the energy in this reg;on of the spectrum transitions from the unvoiced synthesis algorithm to the voiced synthesis algorithm.
s",t(n) = ws(n - N) Mt(0) cos(;:~(0)(n - N) l + ~t(0)~ for 0 < n < N (100) Otherwise, if the l'th spectral amplitude is voiced for both the current and the previous frame, and if ~wo(0)-~0(-1)) >_ .1 wo(0), then 3",t{n) is given by the following equation. In this case the energy in this region of the spectrum is completely synthesized by the voiced synthesis algorithm.
so.t(n) - ~S(n) Mt(-1) cos(~'ao{-1) n I + ~t{-1)~
+ ws{n - N) Mt(0) cos(~:ro(0)(n - N)! + ~t{0)J (101) The variable n is restricted to the range 0 <- n < N. The synthesis window ws(n) used in Equations (99), (100) and (101) is assumed to be equal to zero outside the range -105 <
n < 105.
A final rule is used if the !'th spectral amplitude is voiced for both the current and the previous frame, sad if y:ro(0) - ~(-1)~ < .1 c:~(0). In this case s"~(n) is given by the following equation, and the energy in this region of the spectrum is completely synthesized by the voiced synthesis algorithm.
3".t(n) = at(n) cos(8t(n)) for 0 < n < N (102) The amplitude function at(n) ie given by, at(n) _ Mt(-1)+ N(Mt(0)- a!t(-1)j (103) SUBST1TUT~ S~E~

_ 78 and the phase function Br(n) is given by Equations (104) through (106).
er(n> = mr(-1) + (~(-1) ~ l + ~~,(o)ln + (~o(o) - ~0(-1)1 ' 2 v (104) ~mr(0) = mr(o) - mr(-1) - (~0(-1) + ~o(o)) ~ l2 (l05) ~c~,(0) = N ~O~r(fl) - 2xt0~r20X + xJ, (106) The phase parameter ~r which is used in the above equations must be updated each frame using Equations (107) through (109). The notation dr(0) and tlrr(0) refers to the parameter values in the current frame, while fir(-1) and yr(-1) denotes their counterparts in the previous frame.
~r(0) = wr(-1) -~ ('~0(-1) +~0(0)) ~ l2 for 1 < l < 51 (107) ~yr(0) for 1 < ! < lid (0) = y(0)+ L°"Llol~ o for ~i~ < l < max(L(-1),L(O)J (108) The parameter L""(0) is equal to the number of unvoiced spectral amplitudes in the current frame, and the parameter pr(0) used in equation (108) is defined to be a random number which is uniformly distributed in the interval (-~, x). This random number can be generated using the following equation, pr(0) 53125 u(l) - ~ (109) where u(!) refers to shifted noise sequence for the current frame, which is described in Section 7.2.
Note that ~r(0) must be updated every frame using Equation ( 107) for 1 < / <
51, regardless of the value of L or the value of the V/UV deasions.
Once a",r(n) is generated for each spectral amplitude the complete voiced component is generated according to equation (95). The synthetic speech signal is then computed by summing the voiced and unvoiced components as shown in equation (110).
3(n) _ b""(a) + ~"(n) for 0 < n < N. (110) This completes the I'.vIBE speech synthesis algorithm.
SUBSTITUTE SHEET

WO 92!10830 ~ ~ PCT/US91109135 ..4lgonthm Delay (ms.
J

Analysis 72.5 Quantization0.0 FEC 0.0 Reconstruction0.0 Synthesis 6.25 Table 12: Breakdown of Algorithmic Delay 8 Additional Notes The total algorithmic delay is 78.75ms. This does not include any processing delay or transmission delay. The break down of the delay is shown in Table 12 Throughout the figures and flow charts, the variable i is equivalent to the variable z.
For example the variable L in Figure 13 refers to the variable L in the text.
This notational discrepancy is a consequence of the graphical software used to produce this document.
SUBSTITUTE SHEET

WO 92/10830 'l ~ ~ ~ (~ ~ ~ PCf/US91/09135 A Variable Initialization L~rtatileln:tial W
!ue P_, 100 P_z 100 E_,1 P) 0 for al!
P

E- z ( 0 for all P ) P

~avg 10000 ~mm 1000 ~mnr 100000 -of -1 .O2T
) .Il,( 1 for all -1 ) !

L(-1) .~5 laf-1) 12 iy(-1) Oforallk ER O

SE ~.p000 u(n) u(-10~) =
314.

i<~,( 0 for all n. -1 n ) or(-1) 0 for all !

c.y ( 0 for all -1 ) I

SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ ~ ~ ~ PCTlUS91109135 B Initial Pitch Estimation Window u~ll n uy ( n i 1,(I n ~ uy n ly a ~
n 1 n n ~ _ n I n -140 0.00'?~9~-1030.019058-.6 0Ø1.6132--l.tO.O~.i09.~:' ; u.iJ93tS6.~
I

-139 0.0030 -10 0.018 - ~ 0.0-1609-1--130.0 ~ -11 ~ 0.093923 ~ 8 ~ ~ -1 ~ .592'?
~ .p -138 0.0033. -1060.019-lag- O.oa --120.0; 0 1 0.09a 1 ; ~ 039 6; 3; 16 ~ I
-s .

-13 0.4036 -1050.020163; I 0.0-18026--110.0 ~ -9 0.09-1383 ~ ~ 5 3 ~ 539 -136 0.003992-1040.020889- 0.0-18995--100Ø -3 0.09.15.
. 8329 6 -135 0.00.1321-1030.02162 - 0.049966-39 0Ø - ~ 0.094 7 . 9105 . -1.

-13-10.004662-1020.0223. - 0.05093.-38 0Ø -6 0.094896 6 . 9868 -133 0.005016-1010.02313 -69 0.051910-3 0.080616~ 3 0.095022 ~ ~

-132 0.005382-1000.023908-6& 0Ø62883-36 0.081350--1 0.09.6125 -131 0.003 ~99 0.02-1692-6 O.O.i3$.66-35 0.082069~.3 0.09.6205 ~ 60 ~

-130 0.0061.62-98 0.02.5-18p-66 0Ø6-1828-3-10.082,' -~? ~ 0.09.326'?
~ 3 -1'290.006.656-9 0.026290-6.30Ø6.3-33 0.083461-1 U.u9.i?~.i ~ ~ 99 -1?3 0.0069 -96 0.02 -6-10Ø36 -32 0.0841330 , 0.09.3308 ~ 3 ~ 105 ~ 68 -12 0.00 -95 0.02. -h3 0Ø3. -31 0.08.1.:381 ! 0.09.519 ~ ~ 403 930 . 36 -136 0.00 -94 0.028 -62 0.058 -30 0.08.3-1262 ~ 0.09.1262 . 3.16 ~ 6-1 ~ O

-125 0.008302-93 0.029609-61 0Ø39662-'?90.0860483 0.096205 -124 0.008. -92 0.030-163-60 0.060621-28 0.0?G6514 0.0951'?5 ~ 1 -123 0.009253-91 0.031315-.690.0615.-1 0.0~ 3 0.093021 5 ~ ~ '?3 -122 0.009..18-90 0.03219 -53 0.062525-26 0Ø304 6 0.094896 -121 0.01025 -$9 0.0330 -.570.063-1-25 0.083;353~ 0.09-1 ~ 7 7 ~ 1 ~ -1 -120 0.010. -88 0.033965-56 0.06.1-110-24 0.0883833 0.0915 . 9 ~ 6 -119 0.01131-1-87 0.03.1861-.550.0653-1.1-23 0.0893949 0.09-1383 -113 0.011362-36 0.035 -.640.0662.-22 0.08988510 0.09-116 ~ 65 2 -11 0.012-123-85 0.036675-53 0.06 -21 0.0903561 1 0.093923 ~ ~ 193 -116 0.01299 -8-10.03. -.320.068106-20 0.0908081'? ' 0.09366Q
~ 393 -113 0.013585-~3 0.038516-51 0.069012-19 0.09123913 0.093385 -11.10.0141 -g2 0.039446-.300.069910-18 0.0916.191-1 0.093081 ~.5 -113 0.01-1. -81 0.040382-~190Ø -1. 0.09203915 0.092 99 0. ~ 55 -112 0.015-125-80 0.041323--180Ø -16 0.092-10816 0.09_'403 16.

-111 0. 016064- 0.012269--1 0.0 -15 0.092 1 ~ f 0.09-?0;39 ~ ~ ~ 2.3-13 ~ .3.3 .110 0.016 - 0.043219-46 0Ø -14 0.0930':1: ? ~ X7.091649 7 16 . 3-10~?

-109 0.017381-~ 0.0441.-1-45 0.0~-1237-13 0.0933Q.619 I 0.091'?39 ~ , , SUBSTITUTE SHEET

WO 92110830 ~ ~ PCTlUS91109135 n u~l(n) n unnl n u'llnl L ~ ~cyin~
n 20 0.090808.i2 0.0681063.1 0.03 1 0.01'?99 7 .193 I6 21 0.09035653 0.06.19385 0.03667.111. 0.01223 22 0.08988.5:y 0.0662 36 0.035 118 0-011362 23 0.08939455 0.0653448 0.03-1861119 0.011314 2a 0.08888336 0.06-L11088 0.033965120 0.010 ~ ~

25 0.088353i 0.063-1789 0.0330 121 0.01025 7 1 a 7 7 26 0.08 .i8 0.06252590 0.03219 122 0.009 7 804 ~ 7 a8 27 0.08723759 0.06157591 0.031325123 0.009253 28 0.08665160 0.06062192 0.030463124 0.008 29 0.08604861 0.05966293 0.029609125 0.008302 30 0.08.142662 0.058 94 0.028 126 0.00 31 0.084 63 0.05 95 0.02 12 0.00 32 0.08413364 O.O.i6 96 0.02 123 0.0069 33 0.0834616.i 0.05.1 9 0.026290129 0.006356 ~ 99 7 34 0.082 66 0.05432898 0.015.185130 0.0061.12 ~ 7 35 0.0820696 0.05335699 0.02-1692131 0.005 ~ 7 60 36 0.0813.1068 0.052383100 0.023908132 0.00.1382 3 0.08061669 O.O.i 101 0.02313 133 0.00.1016 38 0.0 7 0Ø1093102 0.0223 13a 0.00x662 39 0.07910571 0.049966103 0.021627135 0.004321 40 0.078329.2 0.048995104 0.02088913u 0.003992 41 0.07 7 0.048026105 0.02016313 0.003675 7 539 3 ;

42 0.07673774 0.04x039106 0.0194x9138 0.003371 43 0.0 7 0.046094107 0.018 139 0.0030 as 0.07.5095.6 0.04.5132108 0.018058140 0.002.97 45 0.0 17 0.04.11.109 0.017381 7 x25 4 i a6 0.07340878 0.0x3219110 0.016716 47 0.0725a879 0.0x2269111 0.016064 48 0.07167880 0.041323112 0.015425 49 0.0 81 0.0x0382i 0.014 50 0.06991082 0.039446114 0.014185 51 0.06901283 0.038.516115 0.013585 SlJBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~j ~ ~ ~ PCTlUS91I09135 C Pitch Refinement Window n u'R( n u'R( n u'R~ n u'R( n wRy n 1 n ) n ) n ) n i I

-110 0.01-18 - 0.205353-.16O.60a06 -1-10.956-t.18 0.928916 7 3 . 7 .

-109 0.017 - 0.215294--I50.62080 -13 0.9623 19 0.9210 39 i ~ 7 7 7 7 -I

-108 0.020102- 0.225-166-44 0.63-1.190-12 0.96 20 0.91'?868 -10. 0.022995-75 0.235869-43 0.6.18105-11 0.97294021 0.90-1307 -106 0.026081- 0.2-16-19--~20.661638-10 0.9.7 22 0.395-100 7.1 7 7 592 -105 0.029365- 0.25 -~110.675076-9 0.98181723 0.88613 -104 0.032852- 0.268.113--100.688406-8 0.98561024 0.8 7 ~ 6589 -103 0.0365.16- 0.2 -39 0. ~ - 0.98896725 0.866 7 ~ 9689 01616 7 ~ 05 -102 0.0-10-151- 0.29117-33 0. 7 -6 0.99188426 0.8.56.516 -101 0.04-1:~-69 0.302851-3 0. ~ -5 0.9943582 0.8-16033 -100 0.018915-68 0.31-17-36 0. 7 --1 0.99638618 0.83526 -99 0.053.182-67 0.326 -35 0. 7 -3 0.99 29 0.82-1231 ~ 8'? 52986 7 966 -98 0.0582 -66 0.339018-3-10. i -2 0.99909530 0.81'293.1 i -9 0.063303-65 0.3.11425-33 0.. 7 -1 0.999 31 0.301391 7 7 610 7 7.1 -96 0.068563-64 0.363994-32 0. 7 0 1.00000032 0. ~

-93 0.07-1062-63 0.376718-31 0.8013911 0.99977433 0.7 -9-1 0-0 7 -62 0.389188-30 0.8129352 0.9990953.1 0. 7 -93 0.085 -61 0.-102594-29 0.82.12313 0.99 35 0. 7 -92 0.092009-60 0..115 -28 0.8352674 0.99638636 0.7-10390 -91 0.098483-59 0.428978-27 0.8460335 0.99.135837 0.727620 -90 0.105205-58 0.4.12337-26 0.8565166 0.99188438 0.71-1692 -89 0.112176-5 0.455 - 0.866 7 0.9889639 0. 7 i 7 93 25 7 05 7 01616 -88 0.119398-.560.469336-24 0.8765898 0.98561040 0.688406 -d 0.126872-33 0.482955-23 0.88615 9 0.98181741 0.6 -X36 0.13-1596-.i.~0..196640-22 0.895-10010 0.9 -12 0.661638 i 7592 -R5 0.1.12572-53 0.510379-21 0.90430711 0.912940-13 0.6.18105 - 0.150799-.520.524160- 0.91286812 0.96 4-1 0.63.1-190 -83 0.1592 -.510.53.9 -19 0.92107-113 0.9623745 0.62080 -82 0.168001-30 0.551802-18 0.92891614 0.9564 -16 0.607 -81 0.176974--190.563639-I7 0.93638613 0.95017447 0.5932Q4 -RO 0.186192--180.5 -16 0.9.13.1716 0.913-i-13 0.3 - 0.195653--170.59328-1-1.30.95017 17 0.936386-19 0.56.1639 7 .1 SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ s !~ 2 ~ PC'1'/US91/09135 n wR~nl i n u~Rl n ) .i0 0..5.1 1802 ~1 0.168001 51 0..13797133 0.159276 52 0.52-1160g-~ 0.150.99 J3 0.51039 .g5 0. I-X25 ~ 2 ~i4 0..1966x036 0.13-1596 5.i 0..1.929.1,5g 7 0.1268.

56 0.-169336g~3 0.119398 57 0..1.5.579389 0.1121.6 58 O. a a 90 0.10520,1 .i9 0.-1289.89I 0.09R.1g3 60 0.-! 15 92 0.092009 ~ 2 ~

6I 0.-10?.59a93 0.08.W.81 62 0..3,39.188ga 0Ø9801 63 0.3;6.18 9.i 0.07x062 6-! 0.36399-196 0.068563 65 0.3,51-1~~9~ 0.063303 66 0.339018 98 0.05827 ;

67 0.3267.8199 0.053-18Z

68 0.3Ia72~ 100 O.O~I8gip 69 0.3023.31101 O.paa573 r0 0.~?911~ 102 0.0-10a:~1 71 0.27 9689103 0.0365.16 7 2 0.268a 10~ 0.032832 7 3 0. 25 7 3a 7 I
05 0, 029363 0.2-1697 106 0.026081 0.235869 07 0.022993 7 6 0.225x6608 0.020102 ~ 0.215294 09 0.01.397 78 0.205355 10 0.01x873 9 0.19.5653 80 0.186192 81 0.1.69..1 SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ ~ ~ PCT1US91109135 D FIR Low Pass Filter n ~ hLpFln) -10 - .

i -1 -9 -.013 ~ ~
.5 -.00.

.016900 -6 .030659 -.5 -.004.565 -4 -.063 -3 -.051602 .0936 1 .2973.54 0 .394201 1 .'?9 .0936 ~ 1 3 -.051602 4 -.063.23 .5 -.004565 6 .0306.59 .016900 8 -.007430 9 -.0137 -.00217 SUBSTITUTE SHEET

WO 92!10830 ~ ~ ~ ~ !~ ~ ~ PCT/US91/09135 E Mean Prediction Residual Quantizer Levels bz Quanfi:er 6~ Quanfmer Ge~~el LeL~e!

0 6.92639.1 32 0.128.38 1 .5.919538 33 0.089934 2 3.-1392 7 34 0.058608 3 5.066921 35 0.028591 4 4.617 366 36 -0.000 7 4.1 i 3022 37 -0.0317 62 6 3.8-10832 38 -0.061262 3.39126 ~ 39 -0.09334 8 3.369101 .10-0.130866 9 3.13-1. 12 -I -0.1 i 6572 2.91. 2 7 -12-0.22. 996 11 2.69-1890 -13-0.293296 12 2.-1-1.5578 -~-1-0.358191 13 2.2-1 i-113 45 -0.-132165 14 2.066-100 -16-0.512681 1.8902. 4 4 -0.588985 16 1. i 1JJ85 -18-0.6 i 9135 1 1.552308 49 -0.778364 i 18 1.415369 50 -0.885324 19 1.2897.8 51 -1.037573 1.182236 52 -1.241260 21 1.07155 7 53 -1.413629 22 0.94964 7 54 -1.5 7 8331 23 0.83415 7 55 -1. 7 64566 24 0. 7 21151 56 -2.003467 0.61-1959 ~ -2.225895 ~

26 0.52587 7 58 -1.4-12203 2 0.432231 59 -2.869023 ~

28 0.361030 60 -3.108824 29 0.2916-18 61 -3.256825 0.2333 71 62 -3.582653 31 0.1 i 4993 63 -6.3 7 .1048 SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~j 4 ~ ~ PCTlUS91109135 _ 87 F Prediction Residual Block Average Quantization Vectors Quant~:er V-ector 0 -1.1'?13-I-i-l.i -0.i-115110.0241iJ1.868-X291.i.110J6 1010.5 1 -1.307211-1.862232-1.8-151710.1552841..i 3.380802 i Q928 2 -1.332197-1.9.150.51-1.1 0.5506472.9980020.899190 i 05:18 3 -2.230924-1.15536-1-1.016851-0.6.33682.26210-12.8143 4 -1.964839-1.313696-0.68.19860.9083641.1984922.056 -3.043823-1.1 -0.9531941.4807511.3129582.3 7 i ,5119 8825 6 -1.595925-1.9$8-~-1;-0.9255.151.14-11231.5300.11.83J
i 7 96 1 -1..116111-1. i -1.611.5360.3.110981.6028.112.-1-12980 3 -1.36.18-1-1..1237q-1-0.321a~00..110.-100.i691~61.931682 9 -2.'254895-1.11.1062-0.1365490.2913520.6566262.611568 -1.281107-1.-112519-0.38125-10..1386421.9565880.686291 11 -2..526-180-1.:863-13-0.6192 0.1648631.015 3.501542 i 3 i 31 12 -1.53 -1.09991-0.18-19381.5318851.512929-0.288411 i -128 1 13 -1.1'25614-1.283-114-0.18 0.6029650.9011301.09305 ~ 994 7 14 -2.262-127-1.654139-0.4495920. 7 1.9-193511.639521 -3.088813-1.61014.1-0..1981630.9163591.82 2..154-101 16 -1.1$9819-1.1021060.215:1360.4935431.1086380.4 i 17 -1.929109-1.69.15930.-1 -0.0465451.33-40351.865464 i 0 i 3 18 -2.090753-1.2831650.-1913850.51.11842.3309890.03 i 19 -1.6059 -1.6735810.141 0.6776 1.429$421.024925 i 8 120 i 2 -1.151020-1.5148080.05 2.612929-1.0931081.118517 i 629 21 -2.333448-1.1869330.1981481.12 0.842 1.331856 22 -2.-103527-1.6396210.3.94851.6983561.4712250.518122 23 -1.238319-2.1543330.2711801.2719241.0833370.766310 24 -1.384577-1.22.11841.139517-0.2718690.1515120.995641 -1.505020-1.5069090.6398260.1400040.6-196291.582509 26 -1.6$1351-1.095600.5516040.5113380.1817030.929352 2 -2.496549-1.3525521.0372990.8316170.9961120.978113 i 28 -1.271265-1.0895750. i 1.0871440.193-1910.287 918 i 69 29 -1.488115-1.2413801. 6.568130.912-18-0.2909910.391211 -1.539032-1.6932010.3038861.11116 0.2929611.024259 31 -2.25 -2.5330811.024-1131.8583300.3.00511.551898 SUBST1T~TE SH~~

WO 92/10830 ~' ~ ~ ~ ~ ~ ~ PCT/US9l/09135 _ 88 b3 C~uanfmer t~ecfor 32 -1.8 -0. 719680-1.319.1630.1-135361.0 . '? .690.p82 7 0696 5 ~

33 -1.227209-0.882158-2.150-185-0.5-139531.5133..13.2904.

34 -1.109 -0. 7 -0.88.6.11-0.1521.15'?.0906Ø82136-1 35 -2.1610 -1.0 -1.108.5090.1490392.6334991.563531 36 -1.380810-0. 7 -0.99 1.1822331.31225 0.65593.1 7 2-165 i 109 7 3 -1.38508-1.028535-1.2089034.09 -1.319590.34-159-1 7 7 7.16 7 38 -2.224613-1.082153-1.5898172.1 i 2.'?083 0.3108 i 3 7 7 i 4 39 -1.596483-0.895635-1.-1 0.8497582.0021431.111038 i 0780 40 -1.394352-0.862301-0.539571-0.0935.160. 7 2.1116 41 -1.30?. -1.045105-0. x80563-0.4820161.16919 2.-1-11266 -12-2.19 -1.079371-0.-167-120.0415261.2 i 2..1318 7 134 i 0575 7 1 -13-2.226980-0.99 -0..120858-0.8-186171.29-16243.199851 -!-1-1.291955-1.063 -0.6985120..1662801.'?.538241.336518 t 16 -15- 1.61 -0.605015-0.811.1880..189.1540..1681192.0 i i 565 613.1 46 -2.68 -0.892633-0.339170.3221882.1856891..111359 -1 -1.999596-0. 7 -0.-1855810.3865551.2592041.58691-1 r' -17 .18-1.859817-0.6115 -0.1518950. 7 0.4865001.400214 -19-1.163622-0.865261-0.0800000..1132920.1918-151.503 50 -1.-15 -1.009887-0.1208580.226 1.0 7 1.290962 51 -1.243611-1Ø12206-0.02913-1-0.738.080.681 2.376725 ii~1 52 -1.130908-0.974602-0.1151.51.657122O.i2~'ii-t-0.159572 53 -1.284521-1.034056-0.0569561.055 0.8-193040.470063 54 -1.541612-0. i -0.17 0.8180931.55 0.091044 48034 7 285 i .83-1 55 -2.831509-1.081142-0.2032351. 7 1.9661460.3 7 7 .1224 5555 56 -1.643100-0.63 0.8670840.7 7 0.5508620.0892 -1.19 -0.9487210.014473-0.1904280. 7 1.534313 7 7 042 87.1.15 58 -2.035191-0.61 0.2190830.9803571.625333-0.17 x573 1968 59 -2.3 -0.8041840.4075190.6997151.-17 0.5969 60 -2..156881-0. 7 1.1526621.9425410.0 7 0.056590 61 -2..160838-0.622 1.5450551.534188-0. ~ 0. 7 62 -1.2017 -1.0092 0.4 7 1.06 0.910211-0.245421 63 -3.68 -1.0746300.3352131.1.109141.3569461.928823 SUBSTITUTE SHEET

WO 92110830 ~ ~ ~ ~ ~~ ~ ~ PCTIUS91109135 _ 89 _ b3 Quanti_er 6'ector 64 -1..111948-0.26110-0.6.5.5.521-0.12. 1.0415111.~ 1-199.
~ 952 65 -1.348808-0.59094-0.9228 -0. 7 0.-1292393.1-19.131 66 -1.206263-0.39451' -0.3009020.156. 1.2.185Ø6.
i 15 3150 6 -1.33195-0.602382-0.50 -0.28108-11.08-10862.138616 ~ ~ ~ 2-18 68 -1.1652 -0.2. -0.8521850.. 6015Ø6.28490.85.
7 1 3.101 892 69 -1.2-1.1336-0.539809-0.151 0.-1319620.6.129690.861053 i 99 i -1.95 -0..151844-0.1649-150.8-185521.0335000.692328 71 -1.993961-0.-192401-1.6633050.9923951.1 ~ 1.97 9353 ~ 959 7 -1..132 -0..1559110.2285880. 7044610.2601390.695552 2 ~ 90 7 -1.246091-0.260938-0.05 -0.163362-0.0433941.. r i -2.351 -0.330 0.082 0.028 1. i 0.853730 -1 ~ 50 r 30 i 29 i 15 1 i 7 -1.5953310.384.566-0.0394.50-0..1400810.9380301.521-138 x 6 -1.2123.-0.16369Ø29319.1Ø39160.43. -0.328298 i -1.221200-0.1918290.09x.5890. 7 0.-1580450.200 7 .58640 i 95 8 -1.938392-0.1099100.16.18361.9 ~ 0.802510-0.992514 9 -2. i -0.1331810.1342550. ~ 0.9518131.000162 ~

80 -1.1-16268-0.2 0.550-1620.2155 0.2594 0.390833 i 0039 ~ 2 r'9 81 -2.326219-0.137.480..192396-0.3008570.4140681.878401 82 -1.102846-0.3220.100.562617-0.3984:180.9602760.300481 83 -1.842-1-17-0.306 0.353 0.2145820.822 0.928154 i 61 i 2 i 85 i 84 -2.021817-0.58 0.559 1.86 0.531463-0.349226 85 -1. 7 -0.0858590.5816660.2992240.2529310.668821 16 i 86 -1.368053-0.2812210.5124161.2194960. 7 -0.8.8835 87 -1. ~ -0.2 0.54.1.1810.4561250.9 ~ 0.017 23199 7 2-196 ~ 3-18 281 88 -1.375.190-0.217 1.2809540.4004000.1380080.17 89 -1.968945-0.530291.0250010.593126-0.4388911.320046 i 90 -1.332864-0.4351660.602953-0.11 0.3545340.9285 ~ 689 7 3 91 -3.150338-0.-1781810.5932960..24.900.5831851.727288 92 -1.680137-0.433 0.1824292.421587-0.040025-1.050028 ~ 87 93 -1.610010-0.2659861.4733720.908215-0.006118-0..19913.1 94 -3.599 -0.5109170.6228171.302 1.5614730.623620 95 -2.5 -0.2662000.9 7 0. i 0.3481040. i ~ 0812 ~ ~ 88122 23080 SUBSTITUTE SHEET

WO 92!10830 ,~ ~ PCT/US91/09135 b3 Quantt:er Sector 96 -1.-1472762.625359-0.5.13-4-15-0.095192-0.29311-0.2.16233 9 -1.1480090..1 -0.23 -1.063260-0.0 2.0531 7 ~ 5-16 ~ 9 ~ 9359 ~ 3 ~ 7 2 98 -1.656-19.10.22990.1-0.18-1170.1643260.99 0.-1.19063 99 -1.2080Ø133. -0.'?3-10210.066. 0.-1908 0. 7 9 18 90 ~ 0 50762 100 -1.6 -0.05 -0.993-1121. 7 0.94-12160.010284 101 -1.943413-0.012392-0.381380.385 0. 7 1.20 7 7 ~ a3-192 ~ 961 102 -2.5-1-12491.367516-0.-161530.901.1911.332116-0.595296 i 103 -1.301352-0.03194-1-0.2918510.3073951.133356-0.312564 104 -1.1223050.9962600.2058360.150278-0.093496-0.136532 105 -1.299722-0.014 0.0921150.0353390.2043500.982668 106 -1.8482371.05-10290.108145-0.4520280.6460710.492060 107 -2.3566280.1689-150..1569060.2903330..000710.7.10413 108 -1.6 0.2-19 0.117 0.926-1020.200388-0.115919 7 83-12 7 32 . 80 109 -2.5824820. 7145820.20.12 0.8045-180.2095380.6-19580 110 -1.-1265190.-18 -0.151-1.151.2908940.295381-0.495302 111 -1.857239-0.00917-10.0100730.76.12010.3.96100.512570 112 -1.17 0.5.52 0. 7 0.-1822 -0.221-121-0.417 113 -1.3505010..196925O.SJ-1-1830.0821500.13264 0.08433 114 -1.244 0.329-170.882-1-16-0.3421660.596031-0.220954 115 -1.92 0.8 7 0. 5.102170.2353 0.1-118890.021501 7 :195 8511 7 6 116 -1.-1830480. 7 0.8610550.6335440.-18.~?O1-1.2846 117 -1.604940.1605 0.8538130.602.186-0.1-1..i0.136636 118 -1.5458150.023 0.8809490.63 0.633 -0.632329 7 7 7 699 ~ 63 119 -2.3641240.2046980.883-1400.96-14070.6-17.186-0.335966 120 -1.2465121.524-1891. 75-15520..104132-0.2328-16-2.203 7.15 121 -1.32 0.8126161.01 0.732529-0.18 -1.046886 7 389 ~ 086 7 916 122 -2.9124391.0 7 1. 7 0.573 0.311253-0. i 123 -1.1515940.10 1.135731-0.-1630190.0495350.321529 124 -2.229545-0.0661921.9985731..1419700.1111.10-1.255906 125 -1.5552860.8138671.552 1.155 -0. 7 -1.246619 126 -2.5815.160.0946531.0300871.6 7 1.082563-1.303115 12 - 3.35920.26 0.9906300.8801930. 759480.46162 7 7 -1 7 3Z r' 7 SUBSTITUTE SHEET

WO 92/10830 ~ O (~ ~ 4 ~ ~ PGT/US9ll09135 b3 Quantt_er Lector 128 -0. i -1.266712-1.1519 -0.0130581.1900682.130636 58920 r'4 129 -0.670378-1.286163-0.396282-0..1602170.5502912.-1627 q9 130 -0.5 -1.196580-1.281 0.050855i. i 1.256966 7 555 I-1-1 .15JOO

131 -1.02304-1.110196-0.635 -0. J 1. ~ 1.581 7 7 -l I 5 03040 ~ 6~
i ~ .
.1 132 -1.000858-1.323-169-O. i 0.94.1 0. 7 1. ~15 96 7 i 52 18 7 i 63.5 133 -0.-187530-1.58215-1-1.-19J2150.30917 0. 7 2.5324 2 23294 r'3 134 -0.638043-0.63.5886-0.613 0.2261141.0288420.632 7 83 r 96 135 -0.658315-1.625552-1.0573490.65817 1.2597241.423353 136 -0.422482-0.764242-0.104219-0.0382430.4134820.915 137 -0.593983-0.77381.1-0,4-1564Ø53-16.100.2439311.034914 138 -0.432.60-1.103490-0.2234860.07.16131.1849860.50011 i 139 -0.858 -O. -0.4 -0.332 0.8914131..#93440 7 i i 1903 7 -12 i 26 4 r r 6 1-10 -0.8 -0.9. -0.124.521.1039130.-117 0.45 7 8468 5 7 7 -163 i -153 9.5 1.11 -1.093431-2.185463-0.1820511.36 0..1916951.601410 i 880 1-12 -0.823-126-0.583159-0.-1.5333-10.6953920. 7 0.393248 i 3619 i-13 -0.967123-7.023~1.1i-0.213263i.112a932.1230750.968304 144 -0.898506-0.923 0.2353050.1638340.5645500.858652 i 94 1.15 - 0.6549-1.11890.053535-0.2639150.2513191.733048 146 -0.655124-0. -0.0093640.2626121.0700050.057719 ~ 25809 147 -0.888641-0.691490-0.0-14 -0.3746 1.3013-160.698211 148 -0.44 -0.8390990.017 1.350139-0.2268090.145995 ~ 219 033 1-19 -0.8 -0.6802180.5015210.346089-0.2519420.956363 ~ a 150 -0.433120-0.8541510.17 0.8710620.259323-0.016 151 -0.5264 -1.5652580.2=12410.643 0.396-1980.809124 72 ~ 7 31 152 -0.469611-0.8997400.6619210.1735130.2097600.324197 153 -0.802344-0.6833021.373918-0.324524-0.1244580.560749 154 -0.530255-1.1683130.6458100.3854060.944429-0.23 155 -0.615025-0. 0.613220-0.2375150.5992030.342637 156 -0.656610-0.8848821.7279310.533642-0.413702-0.326339 15 -0.852993-0.9795610.8359801.34 -0. 7 0.382152 158 -0.322992-0.9518060.9341231.1279500.220014-0.807248 159 -0.7.1536-0.7282070.5233920.674945-0.0264 0.3017 SUBSTITUTE SHEET

WO 92110830 ~ ~ ~ ~ PCT/US91109135 b3 _Quanci:er l eccor i, 160 -0.823160-0.119285-0.690449-0.0213620.3591891.29510.

161 -0.996040-0.192706-0.985145-0.288130.5366 1.925399 ~ i 0 162 -0.553661-0.3683.16-0.533834-0.013171.513094-0.04-X041 163 -0.8-i365.t-0.5407-1.490688-0.4098651.3213681. 7 16.1 -0.420906-0.181568-0.3834 0.628 0.598312-0.241095 7 4 7 i 165 -0. i -0.188369-0.5220560.6503510.3250090.-17 166 -0.619048-0..109303-0.2319270.4-196121.698979-0.888273 16 -0.639652-0.294231-0.5925640.0240630. 7 0. i i 33291 69132 168 -0.399686-0.08974-~0.1461570.110944-0.0178280.250197 169 -0.588222-0.211102-0.148727-0.112195-0.1291541.250340 1 -0.600610-0.16294-0.0965810.1355810.6968670.02 ~ 7 ~ 7 l -0.659021-0.3609130.1026110.1621-110.0643610.690261 i l 17 -0.806482-0.-1-193530.017 1.1943960.0865.15-0.042363 1 -0.535893-0.16-1005-0.1352920.4353340.2.1-18250.155012 i 17-1 -0.501385-0.5068860.1603070.251-X920.-1856980.110813 115 -0.655061-0.57611-1-0.0743090.5072530.34-167-10.453658 17 -0.693433-0.1322360.3556610.246 0.1.151480.0 7 6 i 30 8170 1 -0.451855-0..126170.-1j 0.0029060.14591 0.211320 1 i 7 8 i 17 -0.-193865-0.1 0.184679-0.17 0.54 0.211666 8 i 8063 5056 7 6 i 9 17 -1.049585-0.1898170.18394 0.1103410.5849690.360185 180 -0.4 -0.3239220.2366281.094113-0.1013 -0.-131370 i 4030 i 8 181 -0. 7 -0.3016130..1057610.73 -0.05.5211-0.04 182 -0.917 -0.516 0.1717401.5664621.18-! -1.48 i 32 7 41 157 i 846 183 -0.874171-0.1285450.1689410.6509960.383389-0.200570 184 -0.960325-0.1509740.6383150.209562-0.1890220.452483 185 -0.632887-0.3618590.521867-0.3413300.1602430.654005 186 -0.661521-0.3788531.230674-0.2420550.290845-0.239045 18 -0.842481-0.5561670.5396420.367 0.4187950.0 i 188 -0.435122-0.0897231.0303080.391201-0.371609-0.525016 189 -0.561265-0.2901460.6655650.38324 -0.3129420.115580 i 190 -0. 7 -0.4097240.6515570.5122430.601696-0.568289 191 -0. i -0.5790511.1146060.3752150.060015-0.217 SUBSTITUTE SHEET

WO 92110830 ~ ~ ~ ~ ~ ~ ~ PCT~US91/09135 63 Quantt.er Lector 192 -0.6431020.13.1225-0.31-16320.2780730.2316910.313805 193 -0.722344-0.042097-0.109619-0.118958O.d917 0.501262 19.1-0.4891160.157 -0.199989-0.2.182020.396386-0.116810 195 -0.9201350.308329-0.16883.-0.5411600.3858950.435948 196 -1.0307590.23655 -0.2345.1.010338-0.1122690.130 19 -0. 7 -0.0858530.150.5380.505988-0.1812830.691858 198 -0.419 0.1596 -0.2631250.6562180.0152 -0.148259 7 42 7 .5 7 3 199 -0.8224530.070949-0.6207720.4364650.7638910.171960 200 -0.4425690.1012280.117 -0.3536850.1759960.401716 201 -0.7027750.1063540.051770-0.161905-0.1251980.831793 202 -0.513167-0.0366310.1135.190.2120590.900156-0.675925 203 -0.9818220.00!4890.193050-0.6891560.5496390.926840 204 -0.4051640.1116190Ø5 0.293954O. l -0.231628 i .122 7 3 205 -1.06 0.1120880.1 i 0.-17 -0.0514800.361859 206 -0.510233-0.0009 0.13 1.3064 0.382524-1.275448 207 -0.480080-0.05 -0.0801840.8 7 0.239609-0.500216 '?08-0.9628880.253-1360.262424-0.1084510.114 0.4-107 i 96 22 209 -0.480026-0.0163 0.3 71212-0.0 -0.2520910.454350 210 -0..1463440.2107810.249658-0.1064460.209687-0.!11296 211 -0.6406580.2150350.399974-0.4758340.4195530.081971 212 -0.4146600.0336-I0.4267330.78237 -0.6-191-0.17 7 7 i 1 8887 213 -0.6885490.28 0.2138560.254846-0.19 0.130247 7.159 7 819 214 -0.8500580.2283840.2956350.743066-0.090860-0.326127 215 -1.063320.02532 0.-X665220.1621990.532233-0.122914 216 -0. 7 -0.0888221.561281-0.069151-0.04;3858-0.632018 217 -1.1002390.29165 0.$13204-0.205241-0.4 0.6 71200 r' 7 0541 218 -0.7683240.0504040. 7 -0.149170.0932160.014785 219 -0.4251730.1158990.606544-1.061910.0233170.741364 i 220 -1.0185400.14.18701.3504690.921043-0.896974-0.500827 221 -0.7-183570.1862310.7353210.175348-0.4348310.086328 222 -0. 7 0.17 0.6284240.6840100.224892-0.959616 223 -0.6621530.0427600.626 0.3157260.099716-0.422714 SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ ~~ ~ ~ PCT/US91/09135 63 Qaanti~er t ector 'I

22.1-0.5986760.942942-0.129648-0.6258520.0920900.319184 225 -0.7241360.499878-0.285218-0.0955860.0903900.514 ~ 13 226 -0..19.1-1580.515644-0..1399550.1615290. ~ -0.50102.1 227 -0.x.194381.094897-0.427-186-0.0228000.110615-0.305808 228 -0.4110450.457059-0.3321970.2-182420.053329-0.0150.18 229 -0.617 0.450654-1.1686481.0182170.2280380.0896 891 ~ 0 230 -0.66 0.3855 -0. 7 0.9185170.81655 -0. 7 7 ~ 7 5 33603 r 19253 231 -1.0811230.556386-0.2894020.-1801650.498430-O.1G-1415 232 -0.6430320.9638660.1013390.431304-0.624722-0.22871.1 233 -0.7940840.5778150.0071410.220547-0.6248150.613436 234 -0.8013910.36-18940.2097840.3 ~ 0..190954-0.641399 235 -0.8613860.338-1100.0622090.1381930.3-11290-0.038677 236 -0.5097.111.66-130.2250301.192333-0.948806-1.623150 237 -0.5030760.5.10-166-0.0264021.018864-0.508410-0.521401 238 -0.8363530.938-1950.0348220. 7 0.509361-1.38487 239 -0.6 0.7 0.2038330.4-101300.0112-15-0.75-1694 7 0521 i 004 240 -0.67.11021.-18527.10.499922-0.398989-0.9951380.083073 2-11-0.893.1211.1516250.663507-1.052168-0.2451960.3 7 242 -0.5368822.1366120.503439-0.462826-0.119841-1.520463 243 -0. 7 0.82021.10.628426-0.3 -0.103230-0.18 24-1-0.4480920.543.1290.41255 -0.004 -0.4311621-0.0664 i 7 54 7 9 245 -0.6376250.4939720.277709-0.207156-0.21~.i550.290697 2-16-0.4972900.4743900.665754-0.0394480.171922-0.775287 247 -0.7322210.4251060.4104500.0927180.071098-0.267111 248 -0. 7 0. 7 1.967998-0.295450-0.655730-1.093686 249 -1.0296561.9285060.969061-0.306823-0.962414-0.598634 250 -1.0421260.9150340.751574-0.51.12610.456422-0.566603 251 -0.701170.7677901.137418-0.246167-0.436290-0.521532 252 -0.5313621.0130950.7317160.833151-0.925267-1.121293 253 -0.-1095160.8907500.670287-0.005746-0.325122-0.820613 25.1-0..1074680.9486840.8870430.271.109-0.032952-1.666677 255 -1.0848800.4388451.2405690.2152830.361943-1.171 ~ 20 SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ 6 ~ ~ ~ PCT/US91109135 _ 95 _ b~ Quantz_er lector '?56-0.362425-~.J3J168-1.16~?6t~9-0.0780130.37.2561.?67139 '1.57-0.219381-0.966841-L.Oi-1506-1.0661.90.3887292.-1382'2 258 -0.'? -0.6.5 -1.1663.110.3203221.5.1-1.536-0.320994 259 -0.3-14968-0.581909-1.36-10180.3-18900l ..1310130.-161023 260 -0.3 ~ -0.649631-1.00.0851.0502180.6059990.3 ~

261 -0.244848-1.25682-1.061 1.48322 0.168 0.911500 7 i 63 7 7 51 262 -0.38-1592-1..1.11821-0.917 1.0225111.28 0..133 . 38 7 997 7 03 263 -0.30 -1..102702-1.65 1.0132631.0 7 1.2 7 7 415 i 923 517 9643 264 -0.2-16309-1.350269-0.7x0341-0.0955161.00075 1..161718 i 265 -0.319622-0.569.05-0.556766-0.0824980.0269131.501.19 266 -0.16.7-16-0.666952-0.60-1323-0.4357821.51'0520.357 r91 26 -0.1 i -1.1461.50-0.836.1840.1.162-181.1892060.819 7 245.1 i 7 ~

268 -0.158. -0. 7 -0.64nn9.i0.44037 0.-1858000.6050.12 28 23.5.18 0 269 -0.094963-1.03 -0. 0.1301500.3258291.4208-1 7 561 i .14259 i 2 -0.231237-1.118 -0.3833910.60505 0.9199390.208447 i i 7 7 0 .5 2x1 -0.360030-1.30903.-0.-19-18381.3260440.642.090.395212 2. -0.20 -1.0-121.180.08-11580.2192210.61 0.328830 2 7 185 i 164 2 -0.06 -0.60016-0.3036360.22904 0.2350990.506754 7 7 057 7 ~

2 -0.090151-0.5198-12-0.0630-11-0.1 0.870370-0.018008 7 i 9288 2 -0.311921-0.60 -0.110323-0.2 0. ; 0.5.17 7 7 987 . 5662 5$323 610 2 -0.3996 -0.. -0.164171.06.1 0. 7 -0.3654 7 7 5 30129 5 i 26 14 7 7 2 2 -0.290638-0.62 0.1-1260.3552350.0 i 0.3.12850 7 7 869 i 7 ~ ~

2 -0.341857-1.290918-0.1852140.694 1.64 -0.524321 7 7 64 7 .186 279 -0.053217-0.980076-0.163950O.Z51-1481.129812-0.18391 i 280 -0.199185-1.2032700.1555210.0356920.013.1981.097 ~ 8a 281 -0.368690-0. 7 0.643133-1.1810030.4423931.232152 6 i 282 -0.161447-0.6812431.2021650.1651770.47 -1.001967 283 -0.284 -0.5894720.556380-0.5845071.021816-0.119382 ~ 95 284 -0.067 -0.9481010.8348260.5591 0.038653-0.416722 ~ 95 ~ 8 285 -0.122922-0. 7 0.9264040.259048-0.297321-0.063689 286 -0.3-11817-0.5008860. 7 0.9133270. 714 -1.506396 21082 i 30 287 -0.219312-0.6789410.4952500.2262360.381493-0.204685 SUBSTITUTE SHEET

WO 92110830 ~ ~ ~ ~ ~ ~ ~ PCT1US91/09135 b3 Quants_er V'ector~

288 -0.244744-0.242663-0.3 -0.0289190.3383030.5-18941 289 -0.193322-0.232346-0.9063690.373402-0.0233580.982033 290 -0.332x95-O.a -O.a8aa070.3549360.3825430.051698 291 -0.139. -0.254208-0.535989-1.059130.8002 1.188893 292 -0.14602-0.092855-0.2818460.3. 0.17 -0.0284-19 7 5 7 3a 293 -0.20988x-0.180847-0.-1592510.715908-0.19477'r0.328891 294 -0.30x433-0.4x0297-o. 7 1.-184208O.:la -0.x.14.186 x245x 7501 295 -0.361467-0..168509-0.2907300.8338980.2132050.073644 296 -0.102727-0.3283960.017357-0.025669-0.1427520.582027 297 -0.18x397-0.x93959-0.1-18780-0.2006170.000x991.027294 298 -0.2 -0.1261-12-0.1-19269-0.0149440..1605820.106578 7 6.5 299 -0.3 -0.331688-O.O.i9562-O..i33855O.a 0.890035 ~ 6562 1167 300 -0.12x-~-O.a30-103-0.0692261.950922-0..167-0.859 301 -0.33961x-0.15 0.0596920.6505 -0.26x8860.051472 719a 7 0 302 -0.237x-18-0.3x03120.02-16160.3609750.1152440.076965 303 -0Ø517-0.251329-0.200 0.317 -0.2560010.4-11945 31 r' 7 936 a 304 -0.059645-0.307581O.a2i89a-0.053890-0.0318150.031077 305 -0.050311-0.1123730.178695-0.206361-0.1001390.290529 306 -0.26919?-0.1016400.27517 -0.1176440.642352-0.429009 307 -0.219013-0.1530860.188492-0.0844720.278266-0.010146 308 -0.094532-0.3899370.0669630.7036060.1161x5-0.412205 309 -0.1-194-0.3'r 0.3128210.276669-0.3100800.245535 310 -0.391826-0.4188540.1563220.4531500.561558-0.360309 311 -0.268501-0.180x990.3132080.2180330.260303-0.342505 3i2 -0.37 -0.x306271.0805530.157-X19-0.7972010.367 313 -0.172212-0.4464380.386169-0.096094-0.4305300.559145 314 -0.3395.9-0.2063910.616071-0.1922540.330043-0.207850 315 -0.385217-0.2881020. 7 -0.183008-0.1415180.203 94098 ~ 88 316 -0.156735-0.3295321.0484441.362695-0.3x4264-1.580569 317 -0.23411.-0.1659121.x181130.552627-0..161012-1.109659 318 -0.336835-0.3084950.64 0.6630030.203 -0.868887 7 x83 7 71 319 -0.2955x3-0.4171860.5320860.3928630.014-194-0.226673 SUBSTITUTE SHEET

PCT/US91l09135 g?
b3 Quantt~er lector 320 -0.2666400.319394-0.583321-0.18.1812-0.1818900.897309 321 -0.333194-0.025204-0.3 -0.93998-!0.23 1.638001 322 .1-1 322 -0.2059 0.0811-13-1.226639-0.5825751.120990.313093 323 -0.23-1935-0.03-1-198-0.684129-O.a 0.2-133821.13.1833 7.1633 324 -0.117. 0.2899 -0.90 0..113-1990.1613600.1608 325 -0.0538590.018273-0.-1927650.173939-0.0539170.408369 326 -0.07 -0.089380-0..131097-0.0935890.966364-0.275180 327 -0.3316860.167477-0.696515-0.0740320.5483430.386453 328 -0.2886560.240834-0.063094-0.04-1409-0.0697660.225131 329 -0.0948160.020667-0.038346-0.400807-0.1300960.6.13437 330 -0.3896050.3-13031-0.22 -0.0433160.384261-0.06 7 01.1 7 316 331 -0.226827-0.069230-0.064963-0.897 1.02-131-10.'?3-1-17 18.1 1 332 -0.259 0.3-1811-0.203.1Ø6331110.300895-0.318828 i 83 i 2 333 -0.35-13110.215911-0.15J6350.3129-10-0.27.06Ø138205 334 -0.17 0.2810 -0.006165-0.0400860..181 -0.5-11230 5263 7 -t i 10 335 -0.155284-0.07-1-!990.072.1730.30192-10..15.1851-0.454480 336 -0.0806640.29.1.3-10.236090-0.1563 -0.3253220.031579 337 -0.2842820.137.x-180.300702-0.118876-0.6681520.663200 338 -0.041.1170.0350620.-11.1086-0.111812-0.036926-0.222953 339 -0.086409-0.0538980.1296870.092859-0.03798-0.028401 3.10-0.188181-0.0292840.1838250.46868 -0.119808-0.305198 341 -0.3533420.0637230.16257.10.409414-0.7035190.423191 342 -0.1848360.1641540.407-1600.1872630.55-1266-1.118268 343 -0.2166050.1975060.3889430.1228950.096942-0.589641 344 -0.1413230.2574821.317236-0.679950-0.18517-t0.031769 345 -0.3448540.2029260.766626-0.395164-0.2897540.060861 3:16-0.2574070.0481941.212209-0.8283680.506946-0.681534 3-17-0.3551690.1884570. 7 -0.068005-0.159397-0.356613 3-18-0.328020-0.0134910.6910641.0017 -0.507665-0.8-13585 3-19-0.2994020.0150970..1386600.21-1339-0.307089-0.061564 350 -0.0488400.3086750.857-1281.0111980.508153-2.636574 351 -0.3708470.3049810.:1956030.405863-0.197073-0.638187 SUBSTITUTE SHEET

_ 98 b3 Quanti:er hector 352 -0.2157 1.811393-0.483980-0.30183-i-0.269943-0.539823 353 -0.290 0. 7 -0.039013-0Ø -0..1.03800.045359 35-1-0.3296300.6804250.00 -0.0 0.096 -0.3 7 205 r 6101 a .~ i 8610 355 -0.15 0.39-16-~8-0.38084-0.510720.30 0.-~0 ~ 050 i 5-14 i 054 356 -0.2881291.07 -0.025-1120.480988-0.26586-0.980.4 920 i 7 35 -0.1355580.350983-0.1.5x0260.625 -0.26 -0..119803 r 9 358 -0.044 0.92 -0.13 0.610 0.28902-1-1.645568 359 -0.3805120.995891-1.0170560.2923440.0160290.093343 360 -0.0446611.1987550.39 -1.040575-0.118787-0.392631 ~ 940 361 -0.1460280.5552290.26-1510-0..196302-0.5271250.349756 362 -0.320 0.536850.1830-13-0.5325590.416903-0.283440 363 -0.238 0.3889030.383-1 -0. 7 0.0355250.1959 ~ 52 ~ 2 65081 ~ 2 364 -0.3358621.342.1940.225941-0..102397-0.301915-0.528322 36.1-0.0732130.6616260.'26 0.026968-0.3 -0.507 7 380 7 -I9 r 42 366 -0.24 0.3662090.1426980.184098-0.195305-0.2-19539 367 -0.1866450.5337 0.126009-0.462821-0.0536590.043385 ~ 1 368 -0.06 1.265-1000.6198170.07-1068-0..517 -1.140262 369 -0.383690..1816990.3335510.051815-0.6705 -0.312738 ~ 71 3 -0.1556290.6499010.818816-0.477 -0.086985-0. 7 371 -0.2893.150..191:1180..1167.11-0.221749-O.OQ3114-0.393911 3 -0.2685080.3219400.41913 1.021412-0.205 -1..198193 ~ 7 . 48 373 -0.1367540.6215330.44'5070.424628-0.912065-0..144809 3 -0.2336580.4128070.94 0.898045-0.190095-1.83-1481 7 7.122 3 -0.2049480.3 0.86-16720.1192220.04. -1.205997 . 7 9359 . 33 3 -0.39 1.9692531.684651-0.436592-0.867189-1.952904 ~ 7180 37 -0.2288461. 7 1. 7 -0.696342-1.529 -1.051500 378 -0.1556211.1996681..150233-0.529325-0.558120-1.-106796 3 -0.2140320.9928331.252521-1.309084-0.025321-0.696877 380 -0.0543720. 7 1.0669690.17 -0.682 -1.227154 381 -0.0681830.8649811.0761420.240819-1.234599-0.8 382 -0.1009081.3619631.3304880.107845-0.428479-2.270870 383 -0.04 0.3820421.19517 -0.285817-0.530926-0. 712854 i 5 3 SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ ~ ~ PCTlUS9l109135 _ 99 _ b~ Quantmer L ector 384 -0.027457-0...0011-0.651659-0.4892.90.29509 1.643368 385 -0.012011-1.32 -0.89898-1.18 0.312 3.412504 ~ 204 ~ ~ 031 ~ 68 386 0.1272 -0.864.154-1.22 ~ 0. 2.0403900. ~
~ 9 ~ ~ 7 ~ 02996 387 0.106992-0.692059-0,942743-0.8131101.27.3551.063606 388 -0.002347-1Ø19509-0. 7 1.54 -0.04 0.28 3611-1 ~ 692 7 651 ~ 969 389 0.17 -0.53100-0. 7 -0.3641670.6817 0. ~
75 71 7 5 ~ 59 93824 390 0.042491-0.875249-1..1624401.554.220.889469-0.1.18953 391 0.244 -0.538175-0.9933680.6291680.6961.13-0.0390 392 -0.002211-0.53049-0.5006190.5082770.4459130.0 ~

393 0.111855-0.550084-0..119303-1.06.515-0.17 2.103016 394 0.236596-1.153.64-0.35068-0.2398241.3908800.116838 ~

395 0.084018-0.8223 -0.6120 0.0551890. ~ 0.539966 ~ 0 ~ 3 5.1310 396 0.033418-0.-1-124-10-0.468-1890.9.106060.096860-0.16991.1 397 0.21.1-11-0.594821-0.4.115710.902638-0..1741350.390789 398 0.134375-0.398985-0.2940611.5960630.545589-1.582941 399 0.117 -0.3 -0.4568380.9196870. 7 -0. 7 680 ~ 7 24385 2 ~
.121 3.13 -1000.193250-0.6890860.1844920.2 7 -0.3002900.339261 401 0.164103-0.-1-489190.1046290.033067-0.6553090.802468 -1020.033197-0.4131930.025699-0.1 0.2842730.249801 r 9 r 38 403 0.156908-1.15 0.19 -0.4087030.60 0.603854 i 3 7 .122 7 Q3 7 ~ ~

404 0.109074-0. 75 0.1964920.5161550.09G.i03-0.161025 i 159 405 0.148929-0.552529-0.084-170.5 7 -0.0890430.006923 -1060.109497-0.5565 -0.0419400.3.186840.55618 -0.-115817 407 -0.019408-1.124397-0.1854290.5564790.2. 0.493706 .1080.122989-0.8623200..185023-0.0716550.0018080.324194 409 0.022907-0.4634550.264744-0.3813630.0613530.495851 410 0.051223-0.6446500.831715-0.3086840.261202-0.190766 411 0.162939-0.5 0.237924-0.0569260.3 7 -0.1-16161 .1120.145912-0.5941621. 7 -0.050569-0.9807 -0.290024 413 0.1.19313-0..19390.5332051.039 -1.2419270.013698 .11.10.171939-0.403630.:1189210.1287260.414817-0. 7 -1150.093459-1.0589540.5101041.046086-0.370186-0.2204 SUBSTf TUTS SHEET

WO 92/10830 ~ ~ PCT/US91/09135 b3 Quantuer Lector .~ -0.016814-0.03-1-~-0.229086-0.09.5360.1165-t60.2591 16 1-1 7 i 6 417 0.211-15$-0.240-192-0.360090-0.1 0.1025680.663391 i 67 9:~

-1180.0168650.03036-1-0.6-!-15600.2062990.-121564-0.030491 -1190.065093-0.16311-1-0.325352-0.393-1830.5035300.313366 420 0.108 0.03 -0.8 1.120-1820.121 -0.512 7 22 7 921 7 6082 ~ 8 7 90 -1210.106964-0Ø0835-0.37 0.5892980.365207-0..113251 422 0.09283 -0.081038-0.5249250.2 i 0. 7 -O.J34758 7 -15 i 33-16 7 i 423 0.190512-0.3-16-1.06-16-l20.2652490.4 7 0.485149 424 0.156185-0.048198-0.011-130.128-X15-0.2843860.059430 425 0.152038-0.360220-0.0451910.1275360.0036800.122196 426 0.07,8666-0.106019-0.092402-0.0706.50.290633-0.100165 -1270.158695-0.3 -0.1. -0.5 0.1808370.. 830 ~ .1305015. 7 8100 7 1 .1280.06993 -0.020-17-0.17 1.O 1 -0.028415-0.8613 7 3 7 38-1 i 7 i 0 -t290.1.15517-0.166842-0.1966360.30-1934-0.3437-12-0-2-13251 -1300.139 -0.193Ø0593050.185 0.069694-0.26067 431 -0.01222.1-0.384992-0.19-120.1693300..1-16684-0.024485 i 3 -1320.18.1311-0.0327220.298817-0.0867-160.236698-0.600328 -1330.20235-1-0.0033100.395358-0.29x3.19-0.5485900.248576 -1340.117 -0.0312160.120893-0..19070.517 -0.234492 .135-0.033648-0.15-1-1500.313 -0.6386490.32.10140.188049 436 0.08.668-0.2409530.2687800.9747 -O.a2~1135-0.668098 437 0.103584-0.15330.33.5330.232091-0.690.1980.17 -138-0.0393 -0.120 0.1007710.6326900.42.5089-0.998432 -1390.166995-0.1839600.2-157 0.-103740-0.0-1$976-0.581535 440 0.030708-0.2 1.104694-0.-115049-0.17 -0.24417 7 56:15 0491 7 441 0.142056-0.10170.5458 -0.03 -0.472664-0.076100 442 0.117 -0.3093.180.58 0.081859-0.053284-0.42-1612 443 0.073490-0.2630920.625654-0..133074-0.1194280.116490 444 0.153333-0.1446290.9480480.085311-0.601747-0..1.10276 445 0.1831510.03-11530.7324100.334014-1.030797-0.252891 -1.160.1.5515-0.0407530.5606080.774495-0.302971-1.166855 44 0.146666-0.3213630.4115670.380003-0.3 -0.238520 SUBSTITUTE SHEET

b3 Quanti_er lector ~-180.1089750.271578-0.073327-0.338309-0.03593.10.067057 4-l90.2 l.i 0.120911-0.1290,13-0.17 -0.3587 0.32433 169 254-1 r'9 7 -150-0.0381380.155987-0.112 - 0.3 0.3446090.02 7 r"-l 7 7 893 -~ 0.1905050.2 -1.080 -0.3234250..1192 0.517 -1520.1923220.-185298-0.07.12360.-131410-0.698 -0.336024 -1530.1118350.061492-0.3222230.-124669-0.239656-0.03607 454 0.09145.10.3-14046-0..1085290.1390710.216 -0.38274-1 -1550.0422760.256099-0.3201600.130133-0.1083630.000055 -1.56-0.0260770.2429550.008957-0.0883480.064569-0.202016 -15 0.1-18 0.0571930.0-16063-0. 7 -0.7584451.254252 458 0.252 0.0501650.038923-0.3073131.098695-1.133135 -1590.2508 0.1-135-11-0.0Z -0.2281510.2061910.3.5585 -1600.1891940.-121708-0.0-1-18140.063755-0.069796-0.560008 -1610.0239360.3309910.017 0.033922-0.5955070.189443 -1620.1882410.2108010.06 0.4 7134.1-0.056 -0.880883 463 0.0088110.159837-0.0011000.2819410.005527-0.454976 464 0.0093480.2431260..1712070.142129-0.413062-0.452709 .1650.1021200.4139910.417288-0.035868-0.9356550.008164 -1660.1886870.2333030.299932-0.1-19922-0.110543-0.461416 467 0.0431000.1557620.321-182-0.-102460-0.104506-0.013339 468 0.2380220.17 0.3.169680.454193-0.580.146-0.638351 469 0.0835440.1405990.1398400-161512-0.331187-0.194269 4 0.2325100.05 0.2984580.5484430.181498-1.31840 -171-0.0205640.5021600.3 7 0.317130-0.05 -1.118624 4 0.00069 0..1710900.54317 -0.386847-0.307434-0.320646 -173-0.0152560.1140460.824198-0. 7 -1.0536220.853312 4 0.1332460.1565200.610148-0.54 0.146868-0.498 4 0.0152170.4407760.973119-0.990254-0.152420-0.286398 -17 0.0169170.17 0.8271840.3 7 -0.5027 -0.889827 -1 0Ø515770.3077670.5806181.041523-1.141679-0.839 ~ 7 66 ~

-1 0.2389890.3285580.524646-0.3286510.443022-1.206524 ~

4 0.0585960.2509790.7.13490-0.1-10676-0.216895-0.695454 SUBSTITUTE SHEET

QuOr111;Er leCtOr 480 0.0667200.33250-0.61~?;1.5~; -0.1.9-191' O.lv'~'020Ø021.

431 -0.0248x90.623946-0.32644~ -0.42.5214-0.'?xa3100.396943 ~

482 0.2014981.3088x0-O.a i -0.81x6250.39x952-1.116611 7 401a x83 0.1; 0.3x-1439~ -0.490:3.~ -0.32'?'?0~0.218013-0.42 ; 913 6 ~ 7 484 0.2012531.1; -0.3'?5061I 0.43.-0.935623-0.536803 3895 I 3.

48.5 0.143x; 0.?62; n7.4290.8'~~ 0.0330.56-0.86x5-s0.249x31 9 03 l 486 0.1600 1.05a -1.x365200.16 0.60826x-0..53 ~ 1 148 ~ ~gax ~ ~

a8 0.0588x90.606125-0.311201-0.0310-0.25 -0.06x928 7 ~ 6 7 7 488 0.2x16931.089326-0.07828x-0.536825-0.602430-0.113xx1 x89 0.2583940. 7 -0.036025-1.069x9-0.005x330.1x33 09225 ~ 7 5 490 0.2324242.1617 -0.002669-1.417 0.022535-0.997 -191 0.20886.1.18.1.5.13-0.262271-0.9axx080.1.56267-0.3-13968 -192 0.0823990. ~ -0.~?6394-1~ 0.:333393-0.008156-0.909025 6.53''?

493 0.03-11420.9?9 -0.1 0.06'?'?62-0.3x37 -0..5537 7 -10 r.~.i86 6x :r~

x94 0.09791-1O..i9.i9~?'?0Ø1.3072-0.016.1270.1903.8-0.9207.9 x95 0.2508090.60.561.5-0.060.114-0.4963030.0193 -0.3189x0 ~ 3 x96 0.22; O.h212540.4.1966-0.131600-0.218955-0.95 215 ~ 7 5x3 -19. -0.0021920.34960?0.0681x3-0.332926-0.199692-0.382900 498 0.0035x81.1506-160.~?68661-0.08566x0.1155~i2-1.552693 499 0.15365 0..5236660.192 -0.408~?330.2. -0. 7 500 0.2-133 1.5686260.3 7 0.036033-0.36'5 -1.658801 501 0.1296530.68011.50.3583261.117 -0. 71.5-1.569281 006 ~ ~

502 0.11x6920.5252 0.4833 0.4 0.203611-1. 7 ~ 6 7 6 71625 98541 503 -0.0286030.5629600.4118760.497819-0.510612-0.933399 .304 0.0338912.18'?00..,81962-O.x -1. lOx -1.217169 ~ 6 7 6004 7 17 505 0.1396631.09x5x30.583909-0.4.32362-0.369059-0.996653 306 0.1809130. 7 0.603303-0.3x62800.428536-1.07 0550x 19x5 .50 0.117 0. 7 0..536801-0. -0.316x65-0.27 508 0.092 1.0282321. 7 0.1.15206-1.09800x-1.916516 509 O. l 0.624x090.5-135100.4243 -1.5 -0.164638 a5 7 ~ 3 7 3365 .51 510 0.09a 0.585 0.916208-0.136950-0.0 - i .380515 .511 0.051;221.1013770.526029-0.334292-0.874x21-0.x70375 SUBSTITUTE' SIiEE'T

WO 92/10830 ~ ~ ~ ~ ~~ ~ ~ PCT/US91/09135 b3 (~uonlt_er lector 512 0.515377-0.6892.18-0.6220620.2134900.59478-1-0.012302 ~ 0.3853 -0.604.113-1.3996 0.0 7 -0.068 1.610152 51.1 0.325825-0.301719-0.6113110.2316030.6815270.1-!0322 315 0.35 -0.918 -1.5398800.2101861.0691290.822298 7 104 ~ 9 516 0.45 -0.950972- I ..1-103031.8864260.0362440.011411 :~ 0.2859 -0.32512-0.507 0.2245840.1436150.17 17 7 3 r 209 8504 518 0.543130-1.211391-1.033660. ~ 0.8121910.14190 ~ 4 7 7 519 0.317101-0.7 -0.5969890.3948930.17.29800.289507 7 ..152 520 0.562451-0.294929-0.172360-0.0145470.082645-0.163220 521 0.446232-0.-198199-0.22624-0.092165-0.1812 0.551693 .522 0..131840-0.675845-0.'?03163-0.2728360.3171$50..102859 .523 0.595986-1.2.56.5.8-0.33.59.13-0.3621000..1073901..151296 52.1 0.488101-0.4 -0.1416930.662291-0.053183-0.4 7 .5617 7 9859 .525 0.3.11281-0.6-13882-0.1~-1.1570.1001240.3523500.024624 526 0.312800-0.662335-0.4126970.3710210.389635-0.-198404 527 0.548618-1.309.128-0.4842600.1844070.5212520.539531 528 0.404817-0.3731350.1453920.115282-0.3940930.101577 529 0.595275-0.6687930.023764-0.629191-0.2127390.891724 530 0.595644-0.531-1660.053046-0.1088080.386002-0.39437 531 0.491360-0.3119360.21.1845-0.209540-0.1231080.136418 532 0.609137-0.68 0.1989821.0829 -0.01 -1.184594 533 0.523935-1.3224-100.1150170.2943230.1206160.268589 534 0.4337.14-0.2924640.1050810.4009150.901492-1.548728 .135 0.292085-1.246073-0.1296781.2831140.400013-0.599420 536 0.585660-0.40 0. 7 -0.018400-0.861484-0.019480 537 0.529495-0.5 0.571212-0.678586-0.3389230.490079 538 0.401348-0-322 0.589641-0.344005-0.091344-0.232844 539 0.53570-~-0.5217 0.420773-0.6754580.303443-0.062659 5-10 0.4404 -0.4317180.4453581.593450-0.850542-1.196982 5.11 0..127155-0.3632330.7767180.275629-0.419699-0.696530 5-12 0.322771-0.6059150.5552450.382679-0.008073-0.846667 543 0.456962-0.3469150.3125610.104994-0.063605-0.263958 SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ ~ ~ PCT/US91/09135 65 Quanti~er t'ector 5.1-10.472565-0.080086-0.3 0.11.17 -0.3310540.098106 7 x251 60 .5450.546526-0.185636-0.517 -O.5n8548-0.2413360.986813 546 0.430623-0.05-1.518-0.396393-0.80.13131.'2696140.056026 3-170.604461-0.138267-0..53.884-O.a.545150.102503O..t237-t2 5-180.3854470.103508-O.Q036350.386566-0.003468-0.068378 549 0.4180350.01980.5-0. 7 0.36480 -0.6006810.53 7 .59392 7 466 550 0.447503-0.130694-0.8 0.8605240. 7 -1.028417 .5510.3157 -0.131440-0.2661330.2884170.045864-0.252440 552 0.4974 -0.225321-0.183125-0.3 0.480985-0.195915 553 0.306322-0.119213-0.114539-O.8JJ7390.3310090.452200 .5540.3109060.099101-0.23.5356-0.22213-10.675517-0.628025 535 0.5335170.019330-0.17 -1.63 0..53830. 7 1940 1825 7 968 i 8 7 556 0.-1359710.06~?:360-0.18.p2100.92231-1-4.592532-0.64886-~

557 0..119239-0.143971-0.206139-0.070794-0.0869130.088618 538 0.403239-0.095861-0.21-14510.3565830.343609-0. 7 339 0.302788-0.133887-0.263832-0.0160000..115283-0.314313 560 0.3532010.0619120.096600-0.184167-0.193248-0.134259 561 0.353819-0.1.158990.053727-0.466950-0.0662610.271605 562 0.562189-0.05 0.0636 -0.1303620.161094-0.600917 563 0.2826 -0.1396350.219001-0.2907290.132 -0.18-1059 7 9 ; 33 564 0.547 -0.0007 0.2208120.137423-0.62'238-0.282343 760 7.1 565 0.4 -0.046130-0.0813560.126529-0.83170.162264 566 0.4668020.116-138-0.0393050.321212-0.41.1346-0.449761 .5670.336733-0.1151640.05 0.368410-0.333 -0.208283 568 0.592095-0.2118551.'1667 -0.354504-0.108989-1.080486 569 0.550 -0.1-16-1200.30-1813-0.163599-0.17 -0.17 0.5906620.1093210.245551-0.5626160.103551-0.486399 371 0.4420130.0791020..196412-0.899151-0.093002-0.025334 5 0.53 - 0.1116921.1622941.329224-1.1683-1. ~
~ 7 511 7 2 .18925 5 0.34.18380.117 1.3859060.09660 -1.065891-0.8 7 5 0.3093 0.0748500..1482080.1-132 0.004182-0.979856 5 0.48396-0.0023060.32688 0.053617-0.231106-0.631019 SUBSTITUTE-SHEET

WO 92/10830 ~ j PCT/US91/09135 b3 Quantt:er t'ector .5.60.5.102060.365.31-0.3-14021-0.3463520.2-15097- -0.460621 .~ 0.-1194290.17 -0.5617-0.159062-0.2019600.325589 7 g.l S 0.44-13880.17 -1.0382100.0319930.546388-0.161219 0.4226590.2-1 -0.5.17-0.-1629510.381111-0.041390 7 ~ 809 198 580 0.4190640.202096-0.3-128250.3 i -0.121168-0.532812 .p685 581 0.5161050.198.502-0..516.1830.04 - 0.05 -0.288014 7 i 7 834 .5820.4105600.316 -0.3.126.171.3862900.14 -1.918451 .5830.3488060.43 -0.533.1010.2838730.299110-0,856193 i 9.16 584 0.4021220.388183-0.001685-0.521681-0.3066710.039172 .1850.609-1820.43.11600.004668-0.902293-0.4591600.312183 .5860.4 0.-100 -0.'?967-0.8651320. 7 -0..143072 58 0..5696240.4'?.5095-0.028306-0.284583-0.286828-0.394962 .588J.5006800..163963-0.010-1-150. 7 -0.62 -1.053056 589 0.4500280.1 ~ -0.084433-0.015069-0.5017 -0.126462 i ~ 52 2q .5900.4450430.14388 -0.1159090.0655150.314125-0.852622 591 0.3325030.1-16$09-0.302698-0.0689940.0011 -0.108 i 9 i 59 .5920.3223520.1~?6.1080.623988-0.319526-0.3887 -0.364406 593 0.3562 0.3 7 0.338513-0.34-1222-0.52 -0.196389 i 9 34913 i 650 594 0.4529110.450.110.2 -0.356 -0.1-15 -0.678513 i 7 7 i 97 i 67 595 0.3 0.19:30910.06344-0.6439010. l -0.099553 i 8112 7 Oq.'345 596 0.2860570..1410250.5071100.058122-0.702921-0.589353 59 0.5866 0.4056020.2 -0.289225-1.4995 0.520619 7 ~ 5 7 5946 7 8 598 0.59 0.2720630.6184230.602215-0.520091-1.570430 .1990.4-12 0.5-182410.03 0.272034-0.339-104-0.961301 i 5 600 0.4230540.4994821.328819-0.002159-0.704618-1.544538 601 0.59 0.50388 2.025813-0.81064-1.494879-0.821309 .5 602 0.5205080.2616760.815855-0.213301-0.508164-0.876534 603 0.2803920.2833830.801623-0.667518-0.602267-0.095573 604 0.519850.1643921.0208590.588219-1.105388-1.187900 60.50.-1350130.3858951.2208910.024566-1.534559-0.5317 606 0.4158-180.15533Ø9481450.2369530.088519-1.844 60 0.5 0.-19 0.6384620.130258-0.680735-1.161053 7 i 1 SUBSTITUTE-SHEET

WO 92/10830 ~ ~ ~ ~ ~ ~ ~ PCT/US91109135 b~ Quanl~_er Lector i 608 0.3113121.016631-0.0396.1-0.29.812-0.200399-0..96021 609 0.5.89191.432452-0.545560-0..18x8-18I -0..167220-0.313.03 610 0.5.533041.2x5826-0.9x -0.81; 0.130502-0.1 ~
26 ~ 918 0998 611 0.6095980.678634-0..,88969-0.981096-0.1728980.x54.71 612 0.2885740.579580-0.0061-130.015:21-0..71612-0.x05880 613 0.5957520.655531-0.093-1850.x2588-1.348604-0.235042 614 0.43 1.139354-0.85.16510. 7 -0.352586-1.117040 7 811 .18152 615 0.4226960.671598-0. i 0.205 -0.350 -0.1 i . .1636 7 91 7 30 -1679 616 0.5874 3.1875950.260692-1.513390-0.640595-1.841732 617 0..18 0.9356520.115513-0.132355-0.803-156-0.602631 618 0.5200340. i 0.1-117 -0.222x62-0.4.16511-0.891960 619 0.,5129310..212870.2'?52x0-0.9.2.01Ø0.19983-0..1361-10 620 0.2358151.3561560.3513340.91x198-0.3-15156-2.362307 621 0.5558830.6100780.2254 0.821428-1.269594-0-94-1228 i 2 622 0.3312770.825-1610.3-159390.225-1-11-0.2761-10-1.451938 623 0.5329540.~329~2-0.000427-0.1087620.015684-1.172381 624 0.5352240.6865 O. i -0.562-168-0.245368-1.1292 625 0.394 0.6439280.619041-0.484358-0.560 -0.612557 718 i 32 626 0.5258861.0435930.6517 -0.367 -0.106445-1.746941 60 81-i 62 7 0.4491141.1985 0.369392-1. -0.131913-0.168222 i 3 716904 628 0.-1.158851.5 0.3 i 0..130511-1.0 -1. 7 7 2335 1208 i 3098 -16800 629 0.3681530. ~ 0.6341x6-0.015 -1.025.185-0. 7 630 0.5 ~ 1.0861360.68 -0.064025-0.9 -1.314531 8571 ~ 193 7 3304 631 0.3041560.8263860.5198510.5334 -0.891 -1.312068 7 7 ~ 62 632 0.5476422.1569181.188226-0.852384-1.252906-1.787456 633 0.5038581.0087511. 7 -1.304 -0.999935-0.944491 63.1 0.4-193771.9311080.927874-1.927218-0.7'8464-0.602637 635 0.3 7 0.80 0.926551-0.985455-0.832609-0.288280 636 0.2902880.9660211.456 0.5 -2.112085-1.171400 63 7 0.3-1370.8 1.201909-0.51-1487-1.126247-0. 7 95 r" 82533 638 0.5336331..1080550.87x0320.3 -0.7 -2..166389 639 0.4261851.-1041500.983494-0.642243-1.046922-1.124624 SUBSTITUTE SHEET

WO 92/10830 , ~ ~ ~ ~ ~ ~ ~ PCT~US91I09135 63 Qunnti~er lector ' 6.10 0.969692-0.2830.58-1.39. -0..1084331.18 -0.067 i i ~ 5-16 91 i 6.11 0. 7 -1.325 -1.2 -0.421850.3402801.968620 6-12 1.028856-0.9680.19-0...9.18-0...3012-11.6.59229-0.190154 6.13 0.89517-1.. -0.9 -0.2819951.336 0. i -1 50621 i 0905 7 09 i 13 i 8 6-1.10.6-13136-0..109350-0..156-141.239423-0.266386-0..191140 6-15 0.926271-0.521115-0.99.2100.645316-0.1~6i610.109539 6-16 0.7 -0..104853- 1..1688980.8539061.256461-0.950593 64 0. 7 -O.51O-153-1.0504380. i 0.380021-0.346224 7 80036 ~17 648 0. 7 -0.4931.12-0.1-13992-0.3000070.0998780.085907 649 0.871809-0.831949-0.664153-0.100347-0.2584390.983118 6.50 1.039966-0.692383-0.5064 -0.1212330.2344030.045760 i 3 6.51 0.610131-0.41-141-0.63i91~1-0.7661760.7507810.:157695 i 6~2 0.931339-0.1.51.16-0..562512-0.03-124.-0.127100-0.055724 6.53 1.0 -0..1 -0.3563 0.316293-0.534 -0.030130 7 9236 i 4 7 5 7 .15 65.1 0.638996-0.115132-0.6068330.2885 0.29 -0.402564 7 3 ~ 000 655 0.812245-1.41.1-11-0..1 0.9460800.0923890.041310 i i 6.568 656 0.689688-0.680370-0.1142110.266137-0.048890-0.112312 65 0.809939-0.133806-0.101-133-0.241827-0.3032 -0.029563 658 0.6 -0.290538-0.0396460.2288550.02 -0.596691 659 0.66713-1.058893-0.113717-0.2970171.255.589-0.552635 660 0.9 -0. ~ -0.1219641.091184-0.72 -0..#23466 7 9.18697.1-18 i ~

661 0.661230-0.50413-1-0.0832650.645631-0.699.198-0.019924 662 0.61865-0.8116-190.0061180.4298720.43.1969-0.6'r 7 i 956 663 0.703999-3.099907-0.1186271.1555970.6535950.705382 664 0.635534-0.2295170.596157-0.246336-0.05 -0.718123 665 0.942464-0.5638920.2 ~ 0.004604-0.5402 -0.113456 666 1.041015-0.3700570.1-18731-0.371846-0.030371-0.411437 66 1.009960-1.1132290.317 0.111903-0.00230-0.3236 ~ 388 7 7 5 668 1.022111-0.4313 0. 7 0.43 -0.3755 -1.433844 669 0.920102-0.3651800.5 7 0.543298-0.967292-0. 7 i 155 08042 6 0.855539-0.466170.3690390.3048800.386506-1.449746 i 8 6 0.613813-0.1577580.1644430.56585-0.373638-0.81267 ~ 7 7 SUBSTITUTE SHEET

WO 92110830 ~ ~ ~'~ t b~ Quanti:.er Lector 672 0.725668-0.097590-0..1609.-0.1922020.250.260.029535 673 0.690421-0.009682-1. la -0.3565910.-102-1-160.4164 6 0. 9614890.18 -0. i -0.9581420. 7 -0.1.14 6 0.6-t590.17 -1.392243-1.2235090.8919290.900353 i ~ 1 7539 6 0.993100.1.1203-0. 7 0.0066840.150791- 0.5 r7 7 7 17 504 i 5075 6 0.828650-0.004282-0. i 0.503401-0.262-182-0.364 678 1.0184190.296418-1.1382040.5580850.215 -0.950413 i 34 679 0.651124-0.016094-0..99833-0.1263820.848650-0.5;r7426 680 1.0841990.244714-0.646700-0.486811-0.109998-0.085363 681 0.6200490.132967-0.211684-0.439865-0.084047-0.017380 682 0.9.i 0.050935-0.256063-0.3280660.371993-0.796141 i 381 683 0.886900-O.Oi5268-0.199852-0.3515600.108212-0.368392 684 1.0012620.108596-0.3332200.669223-0..90376-0.633446 685 0.921-1-L0.089'r -0.3415530.068 -0.601989-0.13628 686 0. 7 0.106114-0.3609930.551998-0.0 -0.95 35495 i 5210 i 364 68 0.9124210.162930-0.2 0.095293-0.3 -0.522947 7 i 53 7 2286 688 0.876307-0.0792730.222068-0.760347-0.257483-0.001233 689 0.646611-0.1006890.1 i -0.349 -0.5199950.150599 690 1.0045 0.23 0.1 i -0.6 -0.123701-0.622031 7 6 7 465 i 638 7 3907 691 0. 7 -0.109468-0.102184-0.845670.336 -0.058267 7 8855 7 i 80 692 0.92318;-0.0829 0.1.16.1-190.059459-0.389196-0.656783 i 6 693 0.6184890.068542-0.116289-0.07.106-0.122462-0.374134 694 0.6145340.2318320.2489910.2930900.17 -1.563411 695 0.8906810.0870320.258552-0.239371-0.084664-0.912190 696 0.6160 0.1056 1.325237-0.746770-0.610864-0.689306 i 1 i 2 69 0.6695660.2914790.7 7 -0.358277-0.97 -0.404869 698 0.753888-0.0233950.364248-0.5605900.355643-1.089755 699 1.0828090.06 0.4495 -1.258 0.249279-0.590347 700 0. 7 0.17 0.4603111.149223-1.157296-1.356384 7 0.9355610.1600790. ~ 0.006063-1.0 -0.821829 1 r' ~ 0.9202610.0848390.417 0.224 -0.437655-1.210012 03 0. 7 0.1966020.3.16051-0.198180-0.558466-0.535751 SUBSTITUTE SHEET

63 Quantt~er Lector i .04 0.6590880.619801~0.30.~205-0.352841-0.7 0.16-13a 05 0.8683810. 7 -o.06a35-1.a -0.968 0.860439 7 1.0517 0.371.165-0.0998.-1.8007 0.862839-0.385366 06 Oa 5 2 7 0 0.3 0..186255-0.216598-0.8302 -0.0 -0.23 7 7 5.108 7 2 7 692 7 826 7 0. 7 0.653940-0..1326540.210281-0.510404-0.68-17 08 636-1.1 66 09 0.6186 0.5380 -0.-130 -0.2865 -0.229138-0.23026 710 0.6228790.42 -0.2494180.6381170.025 -1..16497 711 0.8317 0.316291-1.6437560.2030760.34:981-0.055319 712 0.9903350.450009-0Ø10009-0.257972-0.542081-0.600242 713 0.8339360.-1095990.310583-0.666509-0.473331-0.414238 7 0.9948610..1620620.216336-0.32017-0.382974-1.0 7 r' 0.7231280.658002-0Ø1.1213- O..i 0.0 71942-0.83 7 0. 7 0. i 0.013-1-150. i -0. 7 -0.90.165 16 6.125 18238 i 3-131 646 7 7 0. 7 0.2965010.1186 -0. l2. -0.8 -0.158948 1 -13221 r 2 473 71933 i 7 0. 7 0.-1319670.36.1621-0.08-1608-0.156563-1.308104 18 .1972 19 0.7104070.307676-0.0284170.026654-0.183-1-18-0.832822 7 0.6945930. 7 0.-12.17-0.32386-1.199191-0.30397 7 0.64-17910..1.1590..119616-0.682529-0.8612330.033422 .22 0.8840830.6618240.-151979-0.30257-0.857196-0.838093 7 0. 7 0.3184530. 7 -0. 7 -0.339-159-0.680044 7 0.699980.3339090.-i 0.308452-0.95 -0.802023 2-1 7 l 7 7 6 25 1.0161050. 7 0.3819850.540229-1.986301-0.683871 7 0.9835580.4230180..1026120.1049810.190298-2.104426 7 0.9435380..1584310. ~ 0.112483-0.56 -1.692 7 0.9002930.6659951.309944-0.718063-0. 7 -1.4.18241 729 0.7852340.5882551.060083-0.699642-1.,156351-0.177540 7 0.9263140.57 0.814963-0.627 -0.699993-0.992704 7 0. 7 0.6532060.869952-1.405262-0.323438-0.520065 32 0.7493140.62 1.1286920.599105-1.687224-1.417 7 0. 7 0.6074820.9 7 -0.180494-1.054886-1.129443 33 8023 ~ 144 7 0.6-138270.57 1.2631600.63141'r-0.883675-2.2323 3~1 7 683 7 2 7 0.9911-110.3126870.952902-0.2 -0.528891-1.4 5346 35 7.1332 r"

SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ ~~ ~ ~ PCT/US91/09135 b3 Qunnt~:er 1 ector .36 1.0685-IO.B3~ ~O.a6a.,a3-O.a9225.ii n).38a35.1-0..563.80 3 0.90gg821.52889x-O.x09x -1.x -0.87 0.32.5.529 ~ 11 ~ ; 6626 329 ' 38 0.6.591330.86.163-0..526189-0.3365xa0.2662810.92985a 39 0.9103981.021 -0.606 -1.690'~9~~I 0.3333-100.03192 x28 ~ 5a -t 0.66.52921.3053.51-0.23-1063-0.0 ~ -O..i9.i48-1.016135 0 ~ X919 7 x 0.8109000.90606.5-0.356-111-0.~?5933-0. ~ -0.32362-I
1 ~ ~ ~ ' -t2 0.698601.5x00 -0.1383350.15-17-O.O9J-19J-?. L59519 ~ ~ 1 11 -I3 0.8289310. ~ -0.6-18160.205305-0.08234-1.0 ~ 35-19 7 I ~ 7 -14 1.0355392.1-1 0.102250-0.985304- 0.951-13-1.318209 ~ 201 7 -15 0.; 1..53 0.023990-0.3.36;3-0.997600-0.9.5942;
; 1602 1x.3 -16 0.9631661.3136a 0Ø1-1302-0..17 - 0.501268-1.3.17 0 2 ~ 086 1 a -1 0. ~ 0..?r~.5.521-0Ø5 -0.310388-O.aa6 -0.602823 ~ 32260 7 ~ ~ 93 a8 O. x23920'?.113690.0'280290.0: -1.09a -1..3.13130 '?0; ~ 0I

-19 1.0281.110.91. 0.032x86-0.020166-0. ~ -1.101452 938 .53818 30 1.0x9580.8a -0.1190-12-0.062x91-0.039036-1.6 ~ ~ 931 7 6910 ~.510.; 1.0-1688x-0.0827220.x980.57-0.5-18-187-1.62-1852 .52 0. ~ 1.3961630.331360-0.8608x9-1.131899-0.-164965 33 0.88 0.930x910.162932-0. -1.116540-0.06812 la-1-1 ~ 90161 54 0. ~ 1.26-12 0.132503-0.982095-0.361-18-0.808860 .55 ~ .p 7 ~ 0-1 ..5.i0.9502610.8633690.5-12618-0.916.168-0.6~.i?39-0..6.1951 ~.560.80022.11.16351-10.x84118-0.526132-0.9-1x111-0.973261 0. . 1.1392590.2986680-168122-1.411991-0.952091 a 390 58 0.8976840.9239 0.641481-0.361091-0.53x -1.56 ~ 9 ; 91 7 221 59 1.0593960. 7 0.390139-0. -0.19'? -1.262991 . 8032 ~ ~ 130 ~60 0-64-16001.5661011-16897.-O.ll.lx-18-0.9.16085-2.319105 61 0.810. 0.8816950.890725-1.021031-1.04 -0.5147.
81 ~ 354 5 62 0.80 1.2518661.x58648-0.517 -0. ; -1.'?
~ 7 903 25806 ; x525 63 0.8782201.1.512830.90-1055-0.950605-0.300031-1.682882 6-1 1.Oi 1.6038391.9-13.525O.x -1.690559-:3.332219 1069 2138x 63 0.6980.500.889.5811.18-1059-0.0 -1.612205-1.084317 ~ .5128 .66 0.; 1.1620211-11x6580.93210.-1.x62001-2.501-194 5.17 x8 6 0.9280-130.980 0. 7 0.'> -1.1 -1.705348 ~ 7 64 -18813 10816 ~ 2848 SUBSTITUTE SHEET

WO 92/ 10830 ~ ;~ ~~ ~ , ~ ~ PCT/US91 /09135 b3 Quantmer Lector .68 1.240628-0.219091-0.5.1-101~0.53~0300.391931-0.307990 a69 1.1076. -0.32013.-0.861438-0.88755-0. I08 1Ø0212 0 i T 11 i 1. i -0..~-~..1.~-0. -0,34 0.919906-1.198132 i 35280 1~ i 1 i O
0 i 900 1 T 1.1.2933-0.528415-1.52 -0.592250.-1-X36110.931514 i i 3-19 i 1,282187-0.660683-0.619.5160.31 0.2.6119-0.595856 .2 i 849 T 1.410493-0.2698 -1.1.1 -0.0056000.233381-0.220543 7 7 3 i 81 3 i 7 1.0953.160.0fi -0.9 0.8850 0.945543-2.016299 7 1941 i 1564 i 3 .75 1.112521-0.895552-1.060783-0.1113540.8832600.071948 7 1.64863 0.03 -0.386853-0.279831-0.726254-0.293576 7 1.088833-0.140599-0.433211-0.502507-0.5812530.568836 i T 1.452706-0.:115237-0.138604-0.5617 -0.3144430.0073 T 60 i 8 ~ 1.24. -0.607589-0.5-135.0-0,666770.1276930.:1.12492 .9 191 ~

7 1..105 U.0-15103-0.2-15601-0.112U -0.4322 -U.56088~1 80 i 7 i 8 7 5 i 1.588838-0. r -0..1-113130.0012-18-0.230894-0.13923 81 7 8602 i 82 1.10283-1-0.22-103-0.-i 0.503. 0.-15.1249-1.41889 T 1 T 6 . i 8 i 83 1.290614-0-308208-0.2018880.28-1.1500.020668-1.085597 84 1.-1 -0.2680180.2101300.0158 -0. i -0.664784 i 656J i 3 69 7 85 1.285104-0.06 0.193823-0.381971-0.448416-0.580908 86 1.2330.19-0.52 -0.13669-0.301920.434655-0.701461 7 579 7 ~

i 1.-103526-1.5697.8-0.072206-0.6. 0.42817 0.483347 88 1. i 0.03804 -0.13 0.319268-1.12535-0.87 86655 i 7 i i 650 7 1.379151-0.-161.1730.1153090.206872-1..1210150.187196 a90 1.435510-1. ~ 0.0566530..1946210. ~ -0.900781 88105 021.11 791 1.234667-0.441830-0.1092700.084856-0.153562-0.614822 92 1. ~ -0.0443991.28 -1.121838-0.842665-1.035205 r 1.1927210.0050080.338942-1. i -0.6898400.912541 .94 1.-128953-0.0051010.796293-0.505851-0.220207-1.494047 7 1.1-1 -0.5182390.684338-0..167505-0.448733-0.397523 95 r i 796 1..17 0.0418210.5239750.921710-1.5296fi-1.434024 i 1. i -0.4697281.221655-0.238532-1.064248-1.230613 98 1.6 T 0.0594260.7521630.336306-0.443590-2.380108 i 1.5.19388-0.9918730.544 -0.072220-0.180983-0.849068 99 i 97 SUBSTITUTE SHEET

WO 92/10830 c PCT/US91/09135 ~a~6~~ s~

Quantt~er Lector .'3001.1133 0.-123909-0.308 -0.3.19413-0.032109-0.3018.16 7 i i 8 801 1.3912870.205.185-1.1123 -1.0535680.16 0..101 i -1 i .151 i 59 802 1.2585 0.529991-1.058-161-0.21 0.3161-15-0.$282 i -1 i 93 i 3 r 803 1.33 0.13-1419-0. i -0..1 0.194697-0.631029 i-1-13 8.1606 i 0884 804 1.1238890.32.1566-0..1131981.019312-0.593. -1.-160815 805 1.~2-18960.132296-0.9.69080.09 -0.5 -0.123832 i 960 i -13 806 1.1142080.17 -0.51 -0.0950980.930183-1.609262 4-153 a-1-13 807 1..1060680.-176954-0.93-1339O..t15907-0.39.6806-0.967 808 1..1590570.508394-0.166010-0.538420-0.806214-0.456767 809 1.2995060.156071-0.122942-0.5-15150-0.9165070.129062 810 1.6930190.303393-0.3-1-1.569-0.544582-0.-162293-0.64-X928 811 1.-1902500.110190-0.211 -0.894662-0.222186-0.2717 ~ 68 84 812 1.2533190.368313-0.1861630.30 -0.555153-1.18 813 1.1003410.359210-0.346801-0.38101-1-0.599926-0.131 i 43 814 1.5846970.211893-0.20 0..1946 0.243191-2.32 i 103 7 3 i 311 315 1.7189910.27019-1-0.2665900.067791-0.155798-1.635148 816 1.6116120.5-12995-0.0-13567-0.318131-0.6017 -1.191166 817 1.0862500.3202940.209568-1.2-16123-0.4117 0.0417 818 1.1269 0.15.1-1630.035945-0.3153 0.275-1-19-1.2 819 1.5391.120.15 0.230231-0.63 -0.30 -0.981681 i 958 7 832 i i 820 1.3302830..1589900.23 0.796315-1.06-1190-1. 7 821 1.1898610.256-1010.0212050.116117-1.07 -0.511420 822 1.4511430..1512070.1131871.215563-0.42.1395-2.806664 823 1.317 0.14 -0.099792-0.042880-0.06.1126-1.258712 824 1.3016330.3291170.3 7 -0.232333-0.843961-0.928854 825 1.2094690.1433861.025251-0. 7 -1.060738-0.595249 826 1.6 0.451 0.538691-0. 7 -0.1991.9-1.39 7 51-10i 37 68402 i 9 827 1.1682230.3383960.300066-0.606049-0.759432-0.-1.11164 828 1.5885230.4063281.0623680.447130-1.663600-1.8.10709 829 1.2198950.4991.161.605 -0.161465-1.639130-1.52-1181 i 7 830 1.3888330.4060961.193858-0.109652-0.951404-1.92 i 691 831 1.4688140.3649180.4055390.095 -0.7 -1.562950 SUBSTITUTE SHEET

WO 92/10830 ~ Q ~-~ !~ ~ ~ PCT/US91/09135 Quant~:er hector 832 1.556 0. 7 -0.94.1292-0.840342-0.101023-0.3-X8104 833 1. ~ 1.2308 -1.369024-0.860002-0.681585-0.059125 83-1 1.097-X320.880541-0.192. -1.231.90-0.076. -0.x76612 ~-1 7 7 835 1.4985630.61 -1.153604-1.6120840. 7 -0.1-1608 836 1.37 0.635812-0.540929-0.094166-0.6-19592-0.728154 837 1.1055020.9168 -0.596251-0. 7 -0. 7 0.110465 838 1..1718341.013081-0.4867.2-0.432285-0..161549-1.104269 839 1.0929760.84 -1.3-11954-0.443062-0.4998860.344224 840 1.5510360.8066050.252888-0.791376-0.653561-1.165552 841 1.3 0. 7 0.193561-0.8 -1.141011-0.316923 842 1.3684890. 7 0.041083-1.1289 0.1639 -1.193503 7 $

843 1.2120270.72.1.1270.00-1640-0.7.19.19-1-0..130231-0.762329 84-1 1..1-12220.3~t~9.i90.2871070.0108.5-1.22937.1-1.3717-18 845 1.2526 0.911186-0.1 -0.253102-1.186938-0.5 7 9 X3.52 7 8238 846 1.1269421.068 0.2969950.1-1197-0.223479-2.411182 7 9.i 0 84 1.0992500. 7 0.3115610.016819-0.428804-1. 7 848 1.3689611.1020630.419162-0.341639-1.081391-1.467117 849 1.-1008900.68-15290.37 -1.41464.1-0.382293-0.664731 850 1.6471911.0038680.6x1200-0.551517-0.229116-2.521587 851 1.5255511.0849840..124490-2.5417130.6401.12-1.233425 852 1.7 1.1-194640.6938930.501936-2.160018-1.961830 853 1.2736460.5906280.7254-140.058015-1..1980-1.149622 854 1.495 0. 7 0.6874800.404826-1.065698-2.292099 855 1.3516110.6068920.4115.9-0.280506-1.420836-0.668701 856 1.2041040.8183 1.244 -0.655856-1.298522-1.342778 857 1.45 1.0197390.7 71897-1.196053-1.231567-0.821010 858 1.4432030.8081121.656918-1.081941-0.165656-2.660596 859 1.1 0.8951481.439217-1.594627-0.945648-0.970557 ~ 6507 860 1.3186031.2231100.7818130.250134-1.17 -2.295330 861 1.3843-181.0763031.91.12 -0.542941-1. 7 -2.109713 862 1.1571910.6826561.161939-0.238419-0.171780-2.591547 863 1.3786101.0582731.022645-0.557690-0.836257-2.065540 SUBSTITUTE St~EET

63 Quanti:,er Lector 364 1.2414921.871-18-0.625595-1.162593-0.636326-0.688422 i ~

365 1.5857 1.305392-0.-L.t3373-0.899309-1.105856-0.-t.L2535 2-1 ~

366 1..1329831.321055-0.362012-1.17 0.601993-1.816.123 867 1.1555411.3345-L-0.6.97 -0. 7 -0.082511-1.0066.
i .6 21082 9 368 1.5 718621.52 -1.005050-0.180645-0.923454-0.99034 869 1. 7 1.593613-0.94.5016-0.3.56988-1.935601-0.1-1-182a 8 1.7 7 1.332682-0.496062-0.359219-0.343166-1.9043 8 1.2 718121.294988-0.3.58943-0.67 -0. 7 -0. 7 872 1.6387271.882092-0.256660-0.981724-0.860412-1.421982 8 1.6814661.5805900.199121-1.108421-1.587898-0. 7 8 1.4210 1.3-132360.332800-1..121860-0.793748-0.881466 8;.51.7553361.522461-0.110295-'?.435-L88-0.-L96079-0.235901 3 1.316.5331.4144050.14917 -0.0856 -1.913290-1.181106 3 1.0882201.618465-0.102-L-0.261029-1. i -0.57 8 1.3999281..138 0.0 7 -0.046163-0.930630-1.936065 8 1.6313951.26.1-170.166342-0.9133 -1.07 -1.071135 380 1.3030412.-1 0.3930 -0.562442-1.336208-2.2 ~ 9908 ~ 8 7 i 881 1.-4262261.88 0.621282-1.069117-1.1'r -1.688219 i 5 882 1. 7 1. 7 0.526 -1.0 -0.652858-2.289442 383 1..1626801.3163880.522318-0.85 -0. 749944-1.694025 884 1.5438 1.5 0. 7 -0.399255-1.5 -1.879789 7 4 7 8035 28 7 71.591 885 1.3466252.5318220.49247-L-0.550323-2.060170-1.760287 886 1.3171011.3065300.-12-18301.012877-0.5 -3..182753 887 1.2620731.7782860.6.103190.241-168-1.410.586-2.511520 888 1.25504 2.8.195671.294.166-1.312813-2-089995-2.006232 889 1.3782 1. 7 1.908158-1.029490-2.182709-1.838526 890 1.6408132.0356201.321268-1.6 -1.6854 -1.63 i 4253 7 6 7 933 391 1.13 1.2 0.939115-1.029388-0.995818-1.328346 892 1.5 7 1.5 1.0068531.1-12 -2.1365 -3.166 893 1.1-158111.2483320.969534-0.92 -2.051044-0.385170 894 1.2762581.7261331.3435080.057553-2.006213-2.391201 895 1.17 1.2560941.154121-0.168782-1.408846-2.010755 SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ ~ PCT/US9l/09135 63 Quanta:er Vector 896 4.125070-0.0087 -1.262364-0.523561-0.810142-1.52023-1 897 1.897259-0.419391-0.990281-0.682 0.0521430.143039 ~ 28 898 2.1330260.222484-1. ~ -1.3436961.468056-0.690268 899 2.39-1068-0.1159 -1.296800-O.98O81x0.455131-0.455566 900 2.100510-0.658872-0.9690180.752379-0.2077.18-1.017212 901 2.3617 -0.889 -1.194917-0.242072-0.5050610.4 ~

902 3.093311-1.033662-0. 7 -0.4 1.147 -1.953 x 903 1.944264-1.288838-1.594.102-0.0 1.007-1020.001755 904 3.8327600.058832-0.5 -1.000154-0.539824-1. 7 ~ 0978 80595 905 2.872729-0.141513-0.619675-0.418519-0.533843-1.159139 906 2.544289-0.036138-0..5~3~73-1.2907870.378870-1.012621 907 2.2817 -1.03 -0.625899-0.329230.432615-0. ~
66 ~ 115 7 22090 908 3.390768-0Ø -0.60. 0.109 -1.-17 -1.3.13 1983 416 7 28 . 301 7 56 909 2.15 0.003617-0.6068840.1439 -0.661009-1.03 910 2.901.94-0. 714606-0..400730.009 -0.040414. l.-~

911 1.820990-0.156485-0.6850740.004929-0.097031-0.887288 912 3. . -0.098790.095019-0.. -1.152615-1.880766 913 2.269331-0.3-X4285-0.22.380-0.715528-0.548247-0.433851 914 2.4029330.1666800.036236-1.533048-0.479092-0.593669 915 1.936330-0..149252-0.283496-1.300151-0.14.16360.241245 916 2.5098340.1287340.1207570.045191-0.890-!98-1.913978 917 2.008946-0.32 -0.0728160.571 -0.5 -1.602390 ~ ~ ~ 68 7 ~

918 2.160011-0.1359410.018013-0.057 -0.02617-1.958112 919 1.864819-1.0312840.0262180.4263310.280 -1.563812 920 3.4598390.0852800.739130-1.668957-1.168360-1..6892 921 1.836683-0.0093880.558933-0.528828-1.030192-0.827168 922 2.0809 -0.4485400.820111-1.0352270.694716-2.111993 923 1.814616-0.5125720.209608-0.614769-0.094020-0.802823 924 2.5519380.2491120.7014841.243292-1.867040-2.8 7 925 2.3026020.0336830.159402-0.112255-1.554503-0.828889 926 2.130272-0.1167251.2424700.235151-0.588631-2.902496 927 2.020047-0.1482580.25x -0.213886-0.61715.1-1.29x888 SUBSTITUTE SHEET

63 Quanu.:er Lector 928 4.17 0.696 -1.3.19229- l.a -1.29853x-0. i 929 1.9081660.320051-0.566308-0.860630-0.622691-0.1 r 854 930 3.0151210.929171-0.751327-0.921182-0.093991-2.165751 931 2.4 0.829x86-0.999009-1.08 -0.600134-0.6167a i -1235 i i 1 932 3.1-168970.. -0.6x8. 0.2.697 -0.541382-3.01016 7 6x 62 . 7 933 2.095 0.610467-0.803988-0.268.166-0.996439-0.63 934 2.8891140.4525 -1.183304-0.233698-0.493940-1.430 935 2.02 0..164881-0.831685-0.-110487-0.049097-1.201324 936 2.1234810.368296-0.247107-0.374267-0.838532-1.031830 937 2.5989100.661130-0.198423-1.3-16967-0.868410-0.846200 938 3. 7 0.:180387-0.333628-0. i -0.439819-2.695921 939 2.3-152860..198.32-0..10 -0. i 0.0-12632-1. 7 i 993 6-1383 13934 940 2. i 0.861409-0.-13,1$89-0.110910-1.3-19881-1. i 7 .5; -10455 9x 2.2640660.904325-0.3909100.069286-1.3 -1.269582 1 7 714x 942 2.4-141.120. 7 -0.11.13010.173-100-0.912629-2.363345 i 2 943 1.9143140.900838-0.2629130.2283-10-0.915086-1.865394 944 2.1666850. i 0.-1x4017-1.02 -1.269540-1.070268 56252 i 106 9.151.9828580.132-t060.129289-1.481113-1.17 0.106947 946 2.6204x80. i -0.03.1473-0.7 -0.662103-1.862709 947 2.0239390.555409-0.09 -1.020165-0.4 -0.987393 7 601 7 3.149 948 3.2081380.434-1830.1831170.453737-1.60x263-2.675173 949 3.039 0.5961120.01925 - 0.636341-1.39 -1.621106 7 89 7 ~ 6 950 2.9694090.3455120.3; -0.398220-0.276-104-3.019580 951 3.0892490.4439110.42167 -0.396382-1.006919-2.551496 952 3.6579550.5081880.883075-0.611148-1.411324-3.026106 953 2.8963840.2758630. 7 -0.3863 -1.525836-2.044071 954 3.3.522880.8497160.935115-0.931351-1.035936-3.169792 955 2.2465710.5606251.1-X5299-1.0534-l9-0.787425-2.111581 956 2.6394810.8715610.-1952750.305419-2.146401-2.165294 ~~ 2.301.1370.9036810.837333-0.33-197-1.875485-1.831949 ~~ 2.682170.5442891.4-X0142-0.221718-1. 7 -2.673573 959 1.8 0.5420281.5 ~ 0.053086-1.403660-2.644932 SUBSTITUTE SHEET

WO 92/ 10830 ~ ~ ~ ~ ~ ~ ~ PCT~US91 /09135 b3 Quant:_er hector 960 4.7 7 1.068217-1.038366-1..119018-1. 7 -1.6.16205 961 2.9544561.033362-1.22 -1.168053-1.3917 -0.200332 962 3.3 7 1.31.!641-0.238129-1.17514-1.16 -2.105609 963 3. 7 1.1.1711-1-1.290611-1.289550-0.-132 -1.919836 8.5 7 86 964 3.6724381.552265-0.652563-0.537661-1.555209-2..1.

965 2.955 1.432130-0. 7 -1.098432-1.602130-0.933055 7 -18 .i-1220 966 2.4838501.5 -0. 7 -0. 7 -0. 7 -1.88 7 9965 2.1385 51423 00881 7 086 967 2.2847821.361812-1.525181-1.1643410.083632-1.040664 968 2.645 1.6884 -0.021202-1.455799-1.460130-1.397009 969 2.3051941.0835390.0289 -1.620893-1.5017 -0.294999 9 3.8281360.953918-0.109923-1.648-129-1.435362-1.588300 9 2.9102410.9664480.19404-1-1.855301-0.650588-1.564805 9 3.7631901.1098290.202 -0.6-18028-1.601605-1.826083 9 2.8891 1..10-1011-0.0656 -0.17 -1.699-107-2.354565 7 i 7 7 4 3502 9 2.6186081.56.19960.3 7 -1.120909-1.392512-2. 0.15889 975 2.1336661.196140-0.218173-0.746023-0.933164-1.432407 9 2.8850321.666660.562031-1.103630-2.222239-1.787821 97 2.20715 1..1804700.827939-1.906289-1. 753214-0.856022 978 3.0449531.1580880. 7 -1.043950-1..105596-2.620055 979 1.8114041.2972151.0 7 -0.922577-1.340632-1.9227 980 3.7653490.9538080.9836940.094901-2.681056-3.116657 981 2.24 1.5 0.6809420.080227-1.8533 -2.731271 982 2.2499571.1198890.5188740.067670-0.866993-3.089357 983 2.2124631.2312880.68617 -0.832900-1.810116-1.486869 984 3.4553150.9790451.628999-1.651349-1.496057-2.915913 985 2.2849671.0343071.649740-1.855818-1.2641-10-1.849016 986 2.3405431.115171.306327-0.853984-0.794666-3.113358 987 1.799 1.1025 1.104030-0.6421 -0.892782-1.4 7 7 ~ 0 i 6 713 988 2.3907451.4105321.3901590.060248-1.889828-3.361816 989 2.2425161.5474131.5155410.205357-2.494216-3.016570 990 3.16793 1.5982501.572423-0.431072-1.866514-4.040984 991 2.5694 1.03 1.814842-0..196575-1.166834-3.758094 SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ ~ ~ 3 PCf/US91/09135 Quanti:er hector 992 4.9450482.29.1659-0.4 -2.101096- 2.209839-2..1.;8802 993 3.0068602.662063-0.704841-1.128526-2.091787-1..43730 994 -1.1 1.943836- 0.951512- i .4 -1. ~ -1.841982 i 6254 7 a 91652 90x 99.5 2.-150980?.284153-0.656215-1. 7 -0.86413-1.-132689 996 3.9969652.248996-0.519865-0.229631-1.952456-3.543969 99 2.6029001. 7 -0.7.5.101-1-0.479208-2.042600-1.119291 998 2.1691652.289925-0.431285-0.085994-1.521833-2.419939 999 1.9659642..166 -0.809228-0.923616-1.382431-1.317412 s 63 1000 3.4762312.041858-0.360476-2.238668-1.804746-1.114159 1001 2.3840501.804902-0.036859-1.21 -2.004646-0.929788 ~ 618 1002 2.4546962.554265-0.381800-1.57.1668-0.534113-2.518339 1003 1.83.12162.039.511-0.2961 -1.x6.11-t0-1.259116-0.554261 ~ 1 1004 3.64023 2.251-168-0.05-124-1.17 -2.060528-2.603595 i i 3295 1005 2.92277 2.3-19663-0.002189-0.311891-2.854134-2.304184 1006 2.7 ~ 2.0412 0.162150-1.1-12x50-1.421501-2.416265 ~ 129 ~ ~

1007 2.1581991.833288-0.078376-0. ~ -1.448388-1. 7 1008 3.25 2.7688150.248 -2.190598-2.209375-1.875386 1009 1.9014341.9059.180.698055-1.183 - 2.141496-1.180115 1010 3.4608201.8982840.5 7 -2.441164-1.167080-2.320952 1011 2.6 7 2.1885330.804305-1.691774-2.0 -1.901505 1012 3. 7 2.0640280.65966 -0.180860-2.53-1304-3. 7 5436.1 7 62854 1013 2.3920342.9950410.765224-0.364580-2.-139594-3.348086 1014 2.3329591. 7 0.317 -0.282492-1.234214-2.864381 1015 . 2.0868282.3825650..197031-0.379098-1.900801-2.686485 1016 3.1113892.6015041.045615-1. 7 -2.618584-2.435045 1017 2.1256722.3311851. 7 -1.552071-2.247685-2.382569 1018 1.9 ~ 2. 7 1.660433-1.684196-1.3 -3.346900 1019 3.0553031.9194061.026054-1.603103-1.095874-3.301746 1020 3.8266172.3087841.690387-0.71-1098-3.431032-3.680619 1021 2.1843192.1008211.622589-0.685298-3.097954-2.124438 1022 2.3159232.8632981.988724-0.965408-2.196562-.1.005935 1023 2.725 2.48 1.592042-1.145629-2.464834-3.194 SUBSTITUTE SHEET

WO 92110830 2 ~ ~ ~ ~ ~ ~ PCT~US91109135 G Spectral Amplitude Bit Allocation L ParameterBit EncodingBits L ParameterBit EncodingBits 9 C4.z b4 10 12 Cl,z ~~ 10 9 Cs.z bs 10 12 C~.~ bs 10 9 Cs.z bs 10 12 C3,z bs 3 9 Q 1 b7 12 Ca.~

be 6 12 Cs.Z 68 6 9 Qz 6 5 12 Cs.z bs 9 Q, blo -1 12 Q1 blo 9 QS bl, 3 12 Q2 bll 3 9 Qs blz 3 12 Q3 blz 2 C3.z b, t0 12 Qs bla 1 10 C4.2 bs 10 12 Qs bls 1 10 ~'s,z 66 10 6, 9 13 Cl.z b4 10 10 Cs,z bs 10 10 Q1 68 3 13 Cz.z 10 Q z b9 1 13 C3,Z bs 10 Q3 blo 3 13 C4,z b,- .5 ' 10 Q4 611 2 13 Cs.Z

10 (Zs blz 2 13 Cs.~ b9 S

10 Qs 613 2 13 Cs.3 blo 11 Cz,z b4 10 13 Qz 61z 3 I1 C3,z bs 10 13 Q3 613 2 11 C~ z bs 8 13 Qa 614 1 11 Cs.z 6~ 8 13 Qs bls 0 11 Cs.2 be 6 13 Qs bls 0 11 Q t bg 2 11 Q~ blo -1 11 CI.Z ba 11 Q3 611 3 14 Cz.Z bs 11 Q4 612 2 1-lC3.2 be 11 Qs 613 1 1~ C4.s b_ 11 Qs 614 1 14 Cs.2 be SUBSTITUTE SHEET

WO 92/10830 ~ PCT/US91/09135 L ParameterBit EncodingBits L Parameter~ Bu EncodingBits 1-1Cs.3 bs -1 16 Cs.z blo -1 14 C6.z 610 4 16 Cs.3 bll 1-1C6.3 bl 3 16 r's.z btz 1-1Q, blz 2 16 C5.3 613 14 Qz 613 3 16 Q1 bl, 14 Q3 bl, '~ 16 Qz bls 14 Q, bls 1 16 Q3 bts 14 Qs bt6 0 16 Qy 6m 1 14 Q6 bl 0 16 Qs bla 0 1J C1.2 15 Cz.z hs g 1 Cl.z 6, 6 ~

1.5C3_z 66 6 1 Cz,z 6s 6 ~

1 C~.2 b. j 1 2.3 66 6 J C i 1J Ca,3 68 -~ 1 C3,2 i 1 CS .2 b9 -~ 1 C;.3 ~'6 J i 1 Cs.3 610 -~ l C,.2 b9 J i 16 C6.z bl l 4 l Cs.3 bto i 15 C6.3 btz 3 1. Cs.z 6t1 4 15 Qt 613 2 1. Cs.3 btz 15 Qz 61, ~ 1 C6.z b13 ~

15 Q3 bts 2 17 C6.3 bl, 13 Q, b16 1 1. Q 1 bls 15 Qs 6t; 0 17 Qz bts 1~5Q6 bta 0 1 Q3 bm 2 ~

1 Q, 6ta 1 ~

16 Cl.z b, ~ 1 Qs bls 0 r 16 Cz,z bs ~ 1 Q6 6~ 0 ~

16 C3_z bs 16 C3,3 67 J 18 C,.2 6, 16 C,.2 b8 ~ is C1.3 65 16 C,.3 b9 .1 1~ Cz.z b6 6 SUBSTITUTE SHEET

WO 92/10830 ~ ~ S ~ J PCT/US91/09135 G ParameterBit EncodingBlts L ParameterB=t EncodingBits ' f Cz.3 b-, p 19 Q., bz0 1 18 ~ bs ~ 19 Qs bzl 0 C3.:

1'~ ~-~3.3 ~ -1 19 Qs bzz 0 13 ~'a.~ blo 18 Ca.3 611 3 20 Cl.z 64 5 18 Cs,z 61~ 3 20 C1,3 bs -1 18 ~ 5.3 613 3 20 Cz.~ bs 5 18 Cs.Z 61, 2 20 Cz.3 b-, 18 Cs.3 bls ? 20 C3.~ be 18 Q 1 bls 2 20 C3.3 b9 3 13 Qz bl- 3 20 C.,.~ blo 3 13 Q3 ble 2 20 C4.3 611 2 18 Q., 619 1 20 Cs.Z biz 3 18 Qs 6z0 0 20 Cs.3 613 18 Qs b:l 0 20 Cs.4 614 20 Cs.Z bls 2 19 Cl.z 6, J 2O Cs,3 616 2 19 C1.3 bs ~ 20 C6.1 617 2 19 Cz.: bs .p 20 Q1 ble 2 19 Cz. 3 6- .5 20 z bls Q

19 C3.z 68 ~ 20 Q3 bz0 2 19 C3.3 69 3 20 CQ, bzl 1 19 C.,,~ 610 3 20 Qs bz~ 0 19 C,.3 611 2 20 Q6 b~ 0 19 Cs.z biz 3 19 Cs.3 613 2 21 Cl.~ 64 19 Cs,z 61, 3 21 C1.3 bs -1 19 C~.3 bls 2 21 CZ.~ bs .5 19 Cs.,, bls 2 21 C~.3 67 -!

19 Q1 bl~ 2 21 C3.Z bs 19 Qz ble 3 21 C3.3 69 3 19 Q3 bls 2 21 C',.z blo SUBSTITUTE SHEET

WO 92!10830 ~ ~ ~ ~ ~ ~ ~ PCTlUS91l09135 j. PnrameterBit EncodingBits G PnrnmeterBit EncodcngBits ~ ~

21 C~ btt p 22 Q2 bzt 3 . ,o 21 C, dtz ~? 22 Qa bzz a 21 . bta 3 22 Q, bza 1 t~5 z . 2 22 Qs bza 0 21 Cs bts a , 0 21 Cs.a i s '? 22 Qs bzs 21 Cs,z bts 2 23 C b~ '1 ~

21 Cs.a btr 2 23 ,, bs 3 C

21 Cs.~ bta 1 23 t,s bs '1 C

21 Q, 6ts 2 z,z 21 Q z bzo 3 z.a b 3 21 Qa b2t 2 23 CZ,; a 21 Q4 62z 1 23 Ca,2 69 3 21 Qs bza 0 23 Ca.a bto 3 21 Qs bza 0 23 Ca.o btt 13 C'v,~ bt~ p 22 Ct 6a -1 23 C,.s bta 2 z 22 , bs -1 23 Cv.~ bt, 2 a Ct 22 .
Cz ~1 23 Cs,z bts p z 22 , b- ~ 23 Cs.a bts a 22 , bs 3 23 Cs.v bin 2 Ca z 22 .
Ca 3 23 Cs.z bts '' a 22 , bto 3 23 Cs.a bts o Ca 22 . btt 2 23 Cs,v 6~ 1 C~

22 . biz 2 23 Qt bzt 2 C~

22 , bt~ 2 23 Qs bzz 3 22 Cs bl, 3 23 Q3 bzs 2 Z

22 , bts 2 23 Q~ bz~ 1 Cs a 22 . bts Cs 2 23 Qs bzs 0 a 22 , btT 2 23 Qs bzs 0 Cs.z 22 Cs,a bta 2~ C b4 22 Cs.4 bts 1 2~ t,Z bs '1 C

22 Qt b2o 2 t.a SUBSTITUTE SHEET

WO 92J10830 ~ ~ ~ ~ ~ ~ ~ PCTJlJS91/09135 L ParameterBit EncoamgBits G Parameter~ Btt EncodingBats 24 C1., b6 ~ 25 C,.z b13 p 24 CZ.~ 6,- -~ 25 C,.3 6,~ 2 C,.3 b9 -~ ?~JC4,4 bls '7 24 Cz.4 b9 -1 25 C 5.2 bl6 2 24 C3.z blo 3 25 Cs.3 bt~ '.

24 C3.3 bll 3 25 Cs., bla 2 24 C3.4 blz 3 25 C6.~ 619 3 24 C,,~ 613 2 25 Cs.3 boo 2 24 C,.3 b14 2 25 C6., 6n 1 24 C,., bis 2 25 C6.s bin 1 24 CS.z bts 3 25 Q1 b:3 0 C5.3 bl' 2 25 Q~ b~1 24 Cs., bas 2 25 Q3 bZS 0 24 Cs.z b,s 2 25 Q, 6zs 0 2-~C6.3 bzo 2 25 QS bZ- 0 2-1CS.~ bZt 1 25 Qs 6~ 0 24 Q 1 6~2 0 24 Qz bZ3 0 26 C1.2 b4 4 24 Q3 bin 0 26 C1.3 bs 24 Q, hZS 0 26 C,.~ bs 3 24 Qs his 0 26 Cz.~ br 4 24 Qs bz ~ 0 26 Cz.3 be 4 26 t:2., 69 3 25 Cl.z b~ 4 26 C3.~ 610 3 25 C1.3 bs 4 26 C3.3 bll 3 25 Cl., bs 3 26 C3., 61~ 2 25 Cz.z br 4 26 C~.~ b13 2 25 Cz.3 6a 4 26 C,.3 bl, 2 25 Cz.4 b9 3 26 C,., bls 2 25 C3.z 610 3 26 Cs.z bls 2 25 C3.3 61, 3 26 C5.3 bl; 2 25 C3., bin 3 26 Cs., bla 2 SUBSTITUTE SHEET

WO

-L ParameterBit Encoding ParameterBit Bils Encoding ~ L Btts 26 Cs.s bIS 2 2 Cs.s 62q 1 ~

26 Cs.2 b2o 3 27 Q1 bzs 0 26 Cs.3 bet 1 27 Q2 b 2s 0 :.6 Cs.q b22 1 2 Q3 b ~

26 Cs,s 623 1 2 Qq b 0 ~

26 Qt bZq 0 2. Qs 2s 0 6~

26 Q~ b2s D 2 Qs 6~ D
r 26 Q3 bzs 0 26 Qq b2T 0 28 C,.~ bq 26 Qs b2s 0 28 C1.3 6 s 3 26 Qs b2s 0 28 Cl.q b 3 s '?3 C2.2 b- 3 2 Cl b ~ ' 7 ~

. q 7 2 3 C2.3 68 3 ~ C, b . s 3 23 C2.q b9 3 2. CI b q . s 3 28 C3.2 2 C2 b 7 ~

. - ~ 28 C3.3 61, p 27 C2 b 3 ~

. 8 ~$ C3.q b1~ 2 2. C2 b q . g 3 28 C3.s bt3 2 27 C3 b 3 . to 28 Cq.2 bin 3 2 C3 b 3 . t I 28 Cq.3 b 27 2 Is 2 C3 b q . in 28 Cq.q bls 1 27 Cq b 2 . t3 28 Cq.s bin 1 7 Cq 6 2 2 . 1q 28 Cs.2 bta 3 Cq b 2 q . ts 28 Cs.3 bts 2 27 Cq b 2 s . ts 28 Cs,q b~
2 Cs b 2 . in 28 Cs.s b 2 it 1 7 Cs b 2 . la 28 Cs.2 6Z~
2 Cs b 2 7 q . t9 28 Cs.3 b~
2 Cs 6 2 ~ s . ~0 28 Cs.q bzq 1 2 Cs b 3 ~ z . in 28 2 Cs.s bZS l 7 Cs b 1 , ~~ 28 Q I 6~ 0 2 Cs b ~
7 q . as 23 Q2 b2- 0 SUESTITUTE SHEET

WO 92/10830 ~' ~ ~ ' ~a '~' ~° PCT/US91/09135 L ParameterBit Encoding~ L Parameter~ Bit EncodingBus Bits 23 Q3 bz8 0 29 Qs b3, 0 23 Q, b~ 0 29 Qs bin 0 28 Qs 6~ 0 28 Qs b31 0 30 C',.~ 6, 30 C,.3 bs 3 29 CI,~ 6, 3 30 C,,, bs 2 29 CI,3 bs 3 30 C,.s b; 2 29 Cl.a bs 3 30 C~,2 6g .3 29 Cz.~ 6,- 3 30 C~,3 b9 3 29 C2.3 ba 3 30 C~.a blo 2 '?9Cz., b9 .3 30 C~.; bll 2 29 Cz.s b,o :3 30 C3.2 blz 3 29 C3.2 b, I 3 30 C3.3 bla 2 29 C3.s bl~ '? 30 C3.a b" 2 C3.4 bl3 ~~ 3~ C3.5 b15 2 29 C3.s b" ? 30 C,.~ 6,s 3 29 C~.z 6,s 3 30 C,.3 bl~ 1 29 C,.3 bls 2 30 C,., 618 1 29 C,., 6,,- 1 30 C,,s bls 1 29 C,.s b,e 1 30 Cs,~ b~ 3 29 Cs.2 6,9 3 30 Cs.s bZ~ 2 29 Cs.s bzo 2 30 Cs,a bzz 1 29 Cs., 6Z1 1 30 Cs,s b~ 1 29 Cs.s b2~ 1 30 Cs.z bZa 2 29 Cs.2 623 2 30 Cs,3 bis 1 29 Cs.3 6z, 1 30 Cs,, 6zs 1 29 Cs.a bzs 1 30 Cs.s bz 1 29 Cs.s bis 1 30 Q, bz8 0 29 Q, bzT 0 30 Qz 6~ 0 29 Qz 6~ 0 30 Q3 b~ 0 29 Q3 b~ 0 30 Q, b~l 0 29 Q, b~ 0 30 Qs biz 0 SUBSTITUTE SHEET

30 ~ ~ ~ ~ ~ 2 ~ PCT/US91/09135 G ParameterBat EncodingBitsG Pararne!erBat Encoding' Bits 30 Qs b33 0 31 Qo b3., 0 31 C,.~ b, ~ 32 C,.~ b1 31 C,.3 bs 3 32 C,.3 bs 2 3 C1.1 bs 2 32 C, .1 bs 2 l 31 ~', .s 6, 1 32 C,.s 6, 2 31 Cz.z 68 ~ 32 C~.z be 31 CZ.3 b9 3 32 C~,3 bg 3 31 Cz.1 blo 2 32 C~.a blo 2 31 Cz.s b" 2 32 C~.s bl t ' 31 C3.z bin '? 32 C3.i bin 2 31 C3.3 b13 2 32 C3.3 bt3 31 C3.1 b11 2 32 C~.a bt1 0 31 C3,s bls 2 32 C3,s b,s 2 31 C,.z b,s 3 32 C~.z 6,s 3 31 C1.3 b1T 1 32 C4,3 b1T 1 31 C,,, 618 1 32 C,,, 6,8 1 31 CLS bls 1 32 C,,s bls 1 31 Cs.~ bio 3 32 Cs,z 6~ 3 31 Cs,3 bZl 1 32 Cs.3 bzt 1 31 Cs., bzZ 1 32 Cs.1 bzZ 1 31 Cs,s bz3 1 32 Cs.s bz3 1 31 Cs.~ bz4 2 32 Cs.s bw 1 31 Cs.3 bzs 1 32 Cs.z bzs 2 31 Cs,, 6~ 1 32 Cs,3 6~ 1 31 Cs,s b2T 1 32 Cs.1 b2T 1 31 Cs.s 6z8 1 32 Cs,s bZ8 1 31 Q 1 b~ 0 32 Cs.s bZS 1 31 QZ 6~ 0 32 Q, 6~ 0 31 Q3 b3i 0 32 Qz b31 0 31 Q4 632 0 32 Q3 biz 0 31 Qs b~ 0 32 Q, 6~ 0 SUBSTITUTE SHEET

WO 92/10830 ~ ~9 ~ ~ ~ ~ ~ PCT/US9l/09135 L ParameterBit Encoa~ngBits~ Pa~meterBit EncodsngBits G

3y Qs 6~., 0 33 Q3 6~ 0 32 Qs 6~ 0 33 Q, 63., 0 33 Qs b~ 0 33 C,.~ b~ ~ 33 Qs 6~ 0 33 Ci.s bs 33 C,., b5 2 3~ C, ,z b, 33 Ci.s b- 2 3~ Ct.3 bs 2 33 Cz.z b~ a 3~ C,,, bs 2 33 Cz.3 69 2 34 C, ,s 6~ 2 33 Cz., bio 2 3-~ C2,Z b8 3 33 C~.s bi i 2 3~ CZ.3 bs 33 C~.~ bi: 2 3-~ Cz.a b,o 2 33 C3.s bis 2 3-1 Cz.s bra 2 33 C3., b" 2 3-~ C3.z 6~z 3 33 Cs.s bis ? 3-~ C3.3 bis 33 C,.z bts 3 3~l C~., 33 C,.3 bi- 1 3.~ C3.s bis 1 33 C,., 6~8 1 3~l C~.s 6,6 1 33 C,.s 6,9 1 3-~ C,.z 6i- 3 33 C,.s b~ 1 3~ C,.3 b~8 1 33 Cs.z bZ1 3 34 C,., big 1 33 Cs.3 bii 1 3~ C,.s b~ 1 33 Cs., bzs . 3~ C,.s bn 1 33 ~s,s b~, 1 3~ Cs.~ baz 3 33 Cs.s bZS 1 34 Cs.s b23 33 Cs.z b~ 2 3~ Cs., b2, 1 33 Cs.a bi,- 1 3.~ Cs.s bZs 33 Cs.4 bzs 1 3~ Cs.s b2s 1 33 Cs.s b~ 1 34 Cs,z bi; 1 33 Cs,s 6~ 1 3~ Cs.3 b~ 1 33 Q ~ 63, 0 3~ Cs., b~ 1 33 Q~ bjZ 0 3-~ ~'s.s b3o 1 SUBSTITUTE SNEFT

WO 92/ 10830 ..

L ParameterBit EncodingBits L ParameterBit EncodingBits 3-~ Cs.s bm 1 35 Cs.~ 6~8 1 3'~ Q1 b 0 35 Cs.3 b ~ 1 3-1 Qi b33 0 35 Cs., 6 1 3-1 Q3 b~ 0 35 Cs.s b31 1 Qa b3s 0 35 Cs.s 6 3z 1 3~ Qs b~ 0 35 Q 1 b~ 0 3-~ Qs bm 0 35 Qz 63,, 0 35 Q3 b~ 0 35 C, b 3 z . , 35 Q, b~ 0 35 C, bs 2 . 35 35 Ct b ~ QS 637 0 ~

. 6 35 Qs b~ p 3.i Ci b s . ' 2 35 Cz.z b8 3 36 Ci,z b4 3 3.5 C2.3 b9 2 36 C, .3 bs 2 35 C,.a ,o '? 36 Ct., bs 2 b 35 C~.s bll 2 36 Cl,s 6~ 2 35 C:.s b i s 2 36 Cl .s be 35 C3,z b13 3 36 Cz.z ~ 3 35 C3.3 biq 2 36 C2.3 bl0 35 C3,q bts 2 36 C~,~ b,l 2 35 C3.s bis 1 36 CZ,s 61~ 2 35 C3,s b, ~ 1 36 Cz,s 613 2 35 C4,2 bl8 2 36 C3,~ bl, 3 35 C,.3 b19 I 36 C3.3 bls 3s c,., b~ i 3s c'3., a,s i 35 C,.s bzl 1 36 C3.s bm 1 35 C.,.s bzz 1 36 C3,s bt8 1 35 Cs.~ bz3 3 36 C,.z bl9 2 35 C5.3 621 1 36 C,,3 bzp 35 Cs,, bzs 1 36 C4,, bZ, 1 35 Cs.s bis 1 36 C,,s bZZ 1 35 Cs.s 62,. 1 36 C~,s 6 SUBSTITUTE SHEET

WO 92110830 '~ ~ ~ ~~ ~ ~ PCT/US9l/09135 L ParameterBit Encoding~ L ParameterBrt Encoatng~
Bits Bus I

36 C;.z bza p 3 Ca.z bis ~ ~

36 Cs.3 bzs 1 3v Ca,3 bzo 1 36 Cs..s bzs 1 3 C a.a bzt ~

36 C;,; bz; 1 3 Ca,; 6ZZ 1 ~

36 C;.s bzs 1 3 Ca.s bz~ 1 ~

36 Cs,z bz9 1 37 Cs,z bza ?

36 Cs.s b3o 1 37 C;,3 bZS 1 36 Cs,a b3i 1 37 C;,a bzs 1 36 Cs.s bsz 1 3 C;.s bz; 1 36 Cs,s b33 1 3. Cs.s bZ8 1 36 Q ~ b3a 0 3 Cs. z 6~ 1 36 QZ b3s 0 3 Cs.3 630 1 ~

36 Q3 b~ 0 3 Cs.a b31 1 i 36 Qa 63- 0 37 Cs.s b3a 1 36 Q; b~, 0 3 Cs.s b~ 1 36 Qs bas 0 3 Cs.,- 63,, 0 ~

37 Qt 6~ 0 3 C', ,z 6., 3 3 Q z b~ 0 . .

3 C,.s bs Z 37 Q3 bsr 0 ~

3. Cl,a b5 2 37 Qa b~ 0 37 Cl.s 6-,, 2 37 Q; b~ 0 37 Ci.s be 2 37 Qs bao 0 3 Cz,i b9 3 ~

3. CZ.s bto 2 38 C,.i ba 3 Cz,a b, l 2 38 C1.3 bs 2 .

3 CZ.s 61 Z 2 38 C, ,; 66 2 .

3. C~.s b,3 2 38 C~.s ~ 2 3 C3,z 6, a 3 38 C, ,s b8 2 .

37 C3.g bl; 2 38 Cz,z bg 2 3 C3.a 6~s 1 38 C~.3 6~0 2 3 C3.s bt ~ 1 38 Cz,a bt ~ 2 3. C3.s bra 1 38 Cz.s biz SUBSTITUTE SHEET

WO 92110830 ~ ~ ~ ~ ~ ~ PCT/US91/09135 L ~ ParameterBit EncodingBitsL ; ParameterB:t EncodingBils 3R C'~.s 6t3 2 39 C~., 6s ':

38 C3.~ 61, 3 39 Ci.s b' 2 38 X13.3 6ts 2 39 Cl.c 6s 38 C3., his 1 39 Cz.~ b9 p 38 C3.s big 1 39 Ci.3 610 2 33 C3.s 6ie 1 39 C'~., 611 38 C4.~ 6is 2 39 CZ.s 61Z

38 C,.s 620 1 39 Cz.s 61s 38 C4., 6z 1 1 39 C3.i 61, 3 38 C4,s hm 1 39 C3.a 61s 2 38 C,.s 6zs 1 39 C3., his 38 C'S.2 b,s 3 39 C3.s 6i,- 1 38 Cs.a 6zs 1 39 C3.s 6ia 38 Cs., 6zs 1 39 C,.z 61s 3 38 CS.s bz,- 1 39 C,.s 6zo 38 Cs.s 6,8 1 39 C.,,,, b~l 1 38 Cs.- 6z9 0 39 C~.s 6~~ 1 38 Cs,z b~ 2 39 C'~.s 6z3 1 38 Cs,3 631 1 39 Ca,- 6z, 0 38 Cs., 6sz 1 39 Cs.z 6~s 3 38 Cs,s 6~ 1 39 Cs.s 6zs 38 Cs.s 6s, 1 39 Cs., 6zT 1 38 Cs. bas 0 39 Cs.s 6~ta 1 38 Q 1 63s 0 39 Cs.s his 38 Qz b3T 0 39 Cs.,- b~ 0 38 Q3 6~ 0 39 Cs.z 631 1 38 Q, 63s 0 39 Csa 63z 1 3~8 Qs 6,0 0 39 Cs., 38 Qs b~l 0 39 Cs.s 6a, 39 Cs.s Gas 1 39 C,.~ b, 2 39 Cs.~ 63s 0 39 Cl .s 6s 2 39 Q 1 b3~ 0 SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ a ~ ~ ~ PCT/US91/09135 L ParameterBit Encoamg~ L ParameterBit EncodingBrts Bsts 39 Qz b3s ~ -LOCs.s 630 1 39 Q3 has 0 ~0 Cs.- 631 0 39 Q~ bao 0 -i0C7.z biz 1 39 Qs 6,1 0 ~0 Cs.3 633 L

39 Qs baz 0 -i0CS.~ 634 Cs.s 635 L

-i0Cl.z ba Z 40 Cs.s bas 1 40 C1.3 bs 2 40 Cs.- b3~ 0 -i0Cl.a bs 2 40 Q 1 63s 0 -i0Cl.s b' 2 40 Qz b39 0 -i0C16 be ~~ -i0Q3 610 0 -l0Cz.z b9 Z ~0 Q, bal 0 -i0Cz.3 blo Z -i0Qs baz 0 -10Cz., bll '' -i0Qs bq3 0 40 Cz.s blz .i0Cz.s 613 2 41 Cl.z 6~ 3 40 C3,z 61., 3 41 C1.3 6s 2 C3.3 615 2 '~ C1.1 66 C3,4 616 Z -~ C1 .5 ~' 1 C3.s 617 i ~ Cl .s 68 L

C3.6 618 i ~ C~.2 -i0C3.~ 619 0 41 Cz.3 610 2 -10C4.~ 6~0 2 41 Cz.~ 611 2 40 C4.3 6z1 1 41 Cz.s 61z 2 40 C.,.~ bzz 1 41 Cz.s 613 2 40 C.,,s 6i3 1 41 Cz,,- 614 0 -i0C4.s 6z~ i 41 C3.z 61s 3 -i0C,,.- 6zs 0 -11C3.3 61s 2 40 Cs.z 6zs 3 41 C3.~ 61z 2 C5.3 6~1 1 ~1 C3.5 618 i C5.9 628 i 41 C3.s 619 i -i0Cs.s 6z9 1 41 C3,- bzo 0 SUBSTITUTE SHEET

L ~ PercmeterBat EncodsngBsts L i Bet EncoringBsts Parameter ~1 Cq.2 6~1 p ~Z C~-3 - bll ~2 ~ C,.3 6ZZ 1 a2 C~ ; blz 2 -~1C~.a bZ3 1 ~2 W .s bs3 p -~ C a.s b~q 1 -!2L z.s alq 2 -~1Cq,6 ba 1 -12C~,- 61s 0 -11~~~.~ bas 0 ~2 C 3.~ bls 3 .~ Cs.i hz- 3 -~2C3,3 61- 2 -~ C5.3 b2a 1 ~2 C3.q b18 1 CS.q b:9 i '~2C3.s b19 1 -~1CS,s 63a 1 -i2C3_6 bZO 1 -~ C.i .6 b3l 1 ~ ~3.' b21 0 ~ C5,- 63Z 0 -12C ~.2 6'1~

~ C5_~ 533 1 ~2 Cq.3 ~3 1 C6.3 b34 1 ~2 C~q.q ~1 -~ C6.q 63s 1 ~2 C,.s ~s 1 c5.~ 636 ~ ~2 e,.6 -~1C6.6 b3' 1 ~2 Cq,, ~T 0 ~ C6,- 638 0 -~2CS.z ~2a ~ Q 1 b39 0 ~2 C5.3 d'19 1 ~ Q~ 6qa 0 a2 Cs,q b~ 1 ~1 Q3 6q1 0 ~2 Cs.s b31 1 ~ Qq b4~ 0 ~2 C 5.6 ~~ 1 ~ Qs bq3 0 42 Cs,T b~ 0 -~ Q6 6qq 0 42 C6. Wq 1 42 C63 b,~ 1 -~2C1.2 bq 3 42 C6.q 63s 1 -~2Cs.3 bs 2 42 C6.s ~T 1 .~2C1,., 66 2 ~2 C6.s bsa 1 ~2 Cl,s 6T 2 42 C6_T 639 0 ~2 Cl.s bs 2 42 Q1 b~ 0 .~2Cl,r bg 0 42 Qz bql p ~2 C~.z blo 3 .~2Q3 bqZ 0 SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ ~ ~ ~ PCT/US91/09135 G ParameterBit EncodingBits~ ParameterBit EncodingBsts C

QI b43 0 ~3 1~5 6 ~~ 1 ~

-~2 Qs 644 0 -~3 Cs..- ~ 0 -~2 Q6 b,s 0 ~3 -~3 Ce.s ~s 1 -~3 Ci.z b, 3 ~3 Cs., ~s 1 C1.3 b5 ~~ -~3 Cfi.S b3' 1 ~3 C,.a bs 2 -i3 CS.s ~3 C,.s b- 2 ~3 Cs.; ~s 0 43 Cl.s ba 2 ~3 Cs.a b,o 0 -~3 C1_- bg 0 .~3 Q1 641 0 -~3 Cz.2 boo 3 -13 Q~ b4z 0 ~3 Cz.3 61 ~ 1 ~3 Q3 643 0 ~3 C~.; blz 2 a3 Q~ 64, 0 ~3 Cz.s bls 2 .~3 Qs b,s 0 -~3 Cz.s bi, 2 -l3 Qs b,s 0 -~3 Cz,- 6,s 0 -13 C3.z bls 3 -~~ Cl.z b4 2 C3.3 bl; 2 -1-~Cl.s bs 2 ~3 C3.a bls 1 4.~ C1.4 bs 2 ~3 C3.s bas 1 .~-~C~.s b~r 2 ~3 C3.s bZO 1 ~ Cl.s be 2 43 C3_; 6z1 0 44 Cl,; b9 0 ~3 ~C4.z 6z~ 2 44 C2.z 610 3 ~3 C4.s bz3 1 ~ C~.3 bll 2 .13 C,., ~4 1 .~-'1Cz.4 blz ~13 C4.s bZS 1 44 Ci.s 613 2 -~3 C.l.s ~s 1 ~ Cz.s b14 2 -~3 C,,- b~7 0 ~ C~,- 6~s 0 -~3 CS.~ bra 2 44 C3.Z bls 3 ~3 Cs~ b29 1 ~~ C3,3 b17 2 C5.4 ~ 1 '~'~C3.4 b 1 a 1 ~3 Cs.s ~1 1 ~~ ~'3.s bls 1 SUBSTITUTE SHEET

WO 92110830 ~ ~ ~ ~ ~ ~ ~ PCT/US91109135 L ParameterBlt EncodingBtts G PnrnfneterBit Encoa~ngBtts i C3. b2o 1 ~5 Ct.s Ir ~>
s , C3.- bz t 0 ~S C l .s bs r'a s b:~ '' ~5 ~- t.' ~ 0 Ca.3 623 1 '~5e2.2 610 Ca.a 6:a 1 ~5 C~.3 btt Ca.s bas 1 ~5 Cz.a btz p Ca.6 626 1 ~J C~.S 613 Ca.t 6~- 0 ~5 Cz.s bla .~4CS.~ bza 2 ~5 Cz,- bls 0 C s.3 bz9 1 ~.iC3.s bls CS.a 630 1 ~v ~3.J bt,-Cs.s 63i 1 ~v C3.a bls C5.6 63Z 1 -~JC3.5 b19 CS..' 633 0 ~J C3.6 620 C5.8 633 1 ~5 C3.' 621 Cg.2 635 1 -~5Ca.2 622 '7 ' C6.3 636 1 -~JCa.3 b23 Cfi.a 63; 1 -~sca.a C6.3 638 1 '~5Ca.s ~5 Cg.6 639 1 ~5 Ca.6 6~6 C6,; 6aD 0 ~5 Ca.7 ~7 0 C6.a bal 0 ~5 CaB bza ~4 Q1 baz 0 -~5Cs.~ bzs Q~ ba3 0 45 Cs.3 630 1 Q3 baa 0 -~5Cs.a b31 Qa bas 0 -~5Cs.s Qs bas 0 ~5 Cs.s Q6 ba,- 0 .~5Cs.; 63a 0 .~5Cs.e bas ~5 CI,z b, 2 ~i5Cs,i 63s '~5C1.3 65 2 ~5 C5.3 63; 1 -~5C1,4 66 2 -~5C5.1 SUBSTITUTE SHEET

~~3~~~~5 L PnrameterBit Encoding~ L Bit E~ncodmgBtts Bats;
Parnmerer ~5 Cs.s b39 1 a6 Ca.4 bxs ~

4.5Cs.s h4o 1 .~6 C4.s bzs a C'~.- bo l 0 46 C4.s bx--~JC-o.8 b42 0 ~6 Via,' b28 45 Q 1 b43 0 -16 C'a.8 bz9 0 -~5Qz h4a 0 46 Cs.~ b3o ~5 Q3 b45 0 46 C 5,3 ~1 45 Q4 b4s 0 .~6 Cs.a b3x -15Qs he' 0 46 Cs.s b33 -~5Qs bas 0 -i6 Cs.s b34 46 Cs.- 633 0 -~6Cl.x ha '' -~6 C's.a bss a6 Ct.3 bs 2 46 Cs.x b3-46 Cl.s bs 2 46 Co.3 b38 46 Cl.s b- 2 46 C5.4 b39 46 Cl.s ba Z -~6 Cs.s bvo 46 C,.,- 69 0 46 Cs.s bin 46 Cx.x blo 2 -t6 Cs.- b,x 0 C2.3 bt 1 2 ~6 C6.8 643 46 Cx.4 blx 2 -~6 Qi bw 0 46 Cx.s 613 2 46 Qx b4s 0 46 Cx.s bt4 2 46 Q3 64s 0 46 Cz,- 61s 0 46 Q, b" 0 46 C3.x bts 3 46 Qs 648 0 46 C3.3 bin 2 46 Qs 649 0 46 C3.4 bta -~6C3.s bts 1 4i C1.2 b, 46 C3.s bxo 1 47 C1.3 bs .~6C3.,- bin 0 47 C,.4 bs 46 C3,s bxx 1 47 Ct.s br 2 46 C4.x bx3 2 47 C,.s bs 46 C,~ 624 1 4~ C,.- b9 0 SU~STtTUTE SHEET

WO 92/10830 ~ ~ ~ ~ ~ ~ ~ PCT/1JS91/09135 G Parameter~ BI( EncodingBtts G ParameterBat Encodingj Btts -k Cz.: b,o ? .~ CS.s b4z ~ ~

Cz,3 bl1 2 a C5,- bq3 0 ~

b12 ~ '~ ~5.8 b44 ~

~-~2 b13 ~ '~ Q t b15 0 ~

Cz.6 614 2 -~~Q~ 64s 0 CZ,- his 0 -t Q3 64~ 0 ~

-~! C2.8 616 1 -~ Q4 b48 r' -~i C3.~ bl; 3 ~~ QS 649 -~ C3.3 618 2 ~! Q6 b50 i -~ C3..i b l 9 i C3.s bzo 1 ~8 C1.2 64 -1 L 3.s bm 1 ~.3C1.3 bs ~

C3,- b,z 0 48 C1.4 bs C3.s bZ3 1 ~8 Cl.s bT

C~.: b,; 2 -18Cl.s ba a C4.3 bas 1 48 Cl,- b9 0 C.I.q bzs 1 48 Cl.s blo 4; C4.s bZT 1 48 C~.z btl Cq.s b:e 1 ~8 C~.3 bm -t; C.,.,- 629 0 .~8Cz.4 bv3 Cq.a b3o 0 48 C~.s 614 Cs.~ b31 2 48 C2.s bls ~ C5.3 b32 1 ~8 Cy T bis 0 i .

4 Cs.4 b33 1 48 Cz,s b1T 1 ~

4 Cs.s bJ4 1 48 C3.z bia ~

~ C5.6 ~5 1 ~8 C3.3 619 i C5.' b38 0 ~8 C3,4 b'10 C5.8 b3T 0 ~g C3.5 ~1 C6.~ b38 ~ -~8CJ.6 b27 ~ Cg.3 b39 1 ~8 C3,T bZ3 0 i -~ C6.4 b10 1 '~$C3.8 ~4 i Cs.s b4l 1 -~8C,.z his SUBSTITUTE SH~EE't ~~~~42~

G ~ ParameterB~t Encoding~ I' j Bu EncoamgBets BotsC
j Parameter -18 C,.3 b26 1 -~9 I ta.- ~ 0 -1~3!'a.a 62,- 1 -19 ~ ~'~ blo 1 s ~3 Ca.s b2s 1 -19 ~'z.: bll , t-a.s 629 1 '~9 r2.3 b12 ~8 t~'.,,- b3o 0 -19 Cz.s b13 0 '~8 Ca.B 631 0 ~9 C2.5 614 0 '~8 Cs.2 632 Z '~9 C2.6 b15 O

-~8 Cs.3 633 1 .~9 Cz,- bls 0 -18 Cs.4 b34 1 ~9 Cz.s bl' 1 ~8 Cs.s bas 1 .19 C3.z bls 3 C5 6 bas 1 -~9 C3.3 bl9 1 ~3 C;,,- b3- 0 -i9 C~ a bzo 1 ~8 Cs.s bas 0 -19 C3.s b21 1 '~8 L 6.2 b39 1 ~9 C3.6 622 1 ~8 C6.3 b,p 1 a9 C3,,- 623 0 -18 CS.a ba i ~ 1 -~9 C3.s 624 i Cs.s 6,2 1 -~9 C,.z 62s 2 C6.6 b43 1 -~9 C4.3 ~'t6 1 -~8 C's.~ b,a 0 -19 C4.4 b2r 1 ~8 Cs,a b,s 0 ~9 C,.s b2s 1 -~8 Q1 b,s 0 ~9 C,.s 629 1 -~8 Q2 64, 0 ~9 C,,- 6~p 0 Q3 bas O ~9 C4.s b3i 0 ~8 Q, bas 0 ~9 Cs.2 632 2 ~8 Qs 6so 0 49 Cs.3 6~ 1 ~8 Q6 b51 0 49 C5.4 634 1 C5.5 b35 1 -19 C,.z 64 2 -~9 Cs.s ~3s 1 ~9 C1.3 65 2 49 Cs,- 63- 0 -~9 C1.4 bs Z ~i9 Cs,a Visa 0 ~9 Cl.s b~ 2 ~9 Cs.2 bas 1 ~9 Cl.s 6g 2 ~9 Cs.3 b,o 1 SUBSTITUTE SHEET

20~~~~~~

L ParameterBit EncodingBUs L Parameter, Brt Encodrng~
Blts C5.4 ~ b4l 1 i0 C 3,- ~2g 0 C6.5 ~4Z I JO C3.8 ~ b24 ~

C6.6 b43 1 ~~ C~ 2 ~:S

-t9 Cs_- 644 0 i0 Ca.3 bzs 1 ~9 C6.a b4s 0 50 C4.4 bz,. 1 C6.9 646 0 J~ C4.5 b28 ~9 Q 1 b4, 0 i0 C4.s bz9 ~9 Qi bag 0 30 C4,r 6~ 0 Q9 b49 0 JO C4.8 b31 Q4 650 0 J0 C5.2 b32 -~9 Q j 5s r 0 i0 Cs.3 63s -~9 Qs 65z 0 i0 C5.4 63a Cj.S b35 .i0 C1.2 6a '-' S0 Cs.s bas 1 50 C,,3 bs ~ 3O Cj,- 63,- 0 50 C1., bs ? ~iU Cj.B ~s 0 C1.5 6' 2 i0 C5.9 ~9 0 :~0 Cl.s bs 2 50 C6,2 b4o J0 Cl, y?g 0 J0 C6.3 611 .i0 Cl.s blo 1 50 C6.4 b4~

50 Cz.z bll 3 50 C6.s b13 50 ~z.3 b1~ 3 50 C5.6 b14 ~0 ~C~.a 613 1 .SO Cs,; 645 0 50 C2.5 b14 1 50 C6,g b46 J0 Cz.6 b1s 1 50 Cs.9 64~ 0 5O C2,- b16 0 50 Q1 64g 0 50 C~.B 61- 1 50 Qz b4s 0 .i0 C3.~ b18 3 .i0 Q3 bso 0 J~ C3.3 619 1 J0 Q4 651 0 J0 C3.4 ~0 ~ 50 Qj b5~ 0 5O C3.5 6z1 1 50 Q6 bs3 0 .50 C3.s bin 1 SUBSTITUTE SHEET

WO 92! 10830 ~ ~ ~ ~ ~ ~ PCT/US91 /09135 L ParameterBtt Encoding L Bits j Bits j I Paromete B
ncodmg ~ C 1.2 64 ~ 31 ~- 5 .5 636 ), ,~ C1.3 6s 3 i C s.6 b3' 1 71 t_ ; ~ b6 1 71 CS..' h38 7 t_ 1 .5 b," 1 J C5,8 b39 .~ C1.6 bg 1 .~ C5,9 b40 JI Cs,- b9 ~ J1 C5.2 b41 51 Cl.s blo 1 .ilC6,3 b42 J C2.2 bl l 3 .7 C6.4 b43 1 J C2.3 612 3 31 r6.s 644 J1 C2.4 613 1 J1 C6.6 615 J C2.5 bl4 I

~2.6 b15 1 '31~ti.8 b4-~ C~.~ b16 ~ J C6.9 b48 0 J C=.8 b!' 1 J Q l b49 J1 C3.2 bl8 3 JI Q2 650 JL C3.3 bl9 1 '~lQ3 b51 J C3.4 b20 ~ J Q, bs 2 il C3.s bzl 1 S1 Qs bs3 0 J1 C3.6 622 1 '~1Q6 654 J C3.' 623 L

J C3.8 b24 .ilC4,2 62s 3 .i Ca.3 b2s 31 C4,, 62~ 1 .i C4.s bz8 l .i C4.6 b29 1 C4 . , b'30 J1 C~.8 631 .i C4.9 63~

J Cs.2 b33 J CS ~ b34 L

~ Cs.4 bas SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ '~ PCT/US91/09135 H Bit Frame Format tlodulator ,ode tt ~ Bct Modulator ~ Code ~ Btt Bat ora .~ umoer Btt lt~ord Number QO Cd ~ Q16 C2 11 Io c, ~? ~? l, 6 c i 11 Q1 ~z 1~ Q1- c, 11 jl C3 1~ h- Cp 1J

Q2 Ci 1'~ Q18 C1 1J

Iz ca ~? 1 I18 cs 1 1 Q3 C1 ~1 Q19 C6 11 I3 c5 1~ I19 c- 3 Q~ ~b ! ~ Qzo co 1-1 L, c- ti Izo e, 1 ~

Qs ~o --'0 Qzl cz 10 Is c 1 ?0 Iz 1 e3 10 Q5 c: 13 Qzz c, 10 I6 c3 13 Izz eo 13 Q7 ~a 13 Qz3 cl 13 h co 19 Iz3 c s 10 QB C1 19 Q2a C5 10 I8 c s 13 Iz, c- 2 Q9 ~6 13 Qzs co 12 I9 c- p Izs c1 12 Qlo ~0 18. Qzs ~z Ilo cl 18 Iz5 c3 9 Q11 ~z 12 QzI ~, 9 hl c3 12 Iz- cs 9 Q 1z ~4 12 Qzs co 11 Ilz co 1 ~ Iz8 cl 11 Q13 Cl 1' Q29 C5 9 I13 c s 12 I~ cz 8 Q 11 C5 1 2 Q~ C3 h, a ,- ~ I~ e, 8 Qls ~0 16 Q31 ~0 10 Its cl 16 l31 cl 10 s.VBS?~TUTE SHEET

ll~od'ulator~ mode Btt .~'umoer~'.loduiator I Btt .~
Btt ~~'orrl i Btt I ~=~xie um~x-r ti'ord ~

Q3~ ~s ~ ~ has ~z 3 132 C5 ? ~ Ii8 ~ C3 Q33 C: ' ~i9 C7 C3 ~ leg ~ C1 ~

Q34 !'0 9 Qso ~., 3 13., c, 9 I;p c s Q35 ~~ ~ Qs~ ~6 3 I35 CS ~ ISI CI Z

Q36 ~6 ~ Qs2 ~0 I~ c2 6 IsZ cl 3 'K 3' CO '~ Q i3 C3 0 I3- C1 ~ I53 C4 Q38 ~3 6 Qs4 c5 0 I3s ca 6 Is4 c6 .o I39 ~' ~ Iss ~ ~ p '.l a0 CO ' QS6 C 2 I40 CI ~ I56 C3 1 Q4i n v QsT c4 Z

I41 e3 .p IsT cs 1 Q 42 C4 ~ S8 CO 1 142 cs a I~ cl 1 I43 a t 6 Is9 ~~ 0 Q44 ~6 ~ Q60 C3 I" e2 ~ l6o c4 0 Q4S C3 '~ Q61 CO 0 I4s ~a -~ Is~ ~i 0 Q 46 CO ~ Q 62 C 5 Ii6 C1 J I5~ C6 Q4: CS .~ Q~ GT I

I4 ;. c6 ~ Ire cT 0 SUBSTITUTE SHEET

WO 92/ 10830 ~ ~ ~ ~ % ~ _~ PCT/US9l /09135 Speech Synthesis Window 11 T1 I ~1' !t n I ~ 1 I
I 1 ( ~ ~ a T1 tL' Ii I U 1W S I
j ~ I
I j I1 12 ~ I
) !t :
~ ~
~

.1050.000000.;a o.62oouo.~.3; l.oooooo.m l.ooooo019 l.ooooo0 ~

-1040.020000- 0.6-10000--i?i 1.000000- 1.000000?0 1.000000 ~ ~ 11 -1030.0-10000- 0.660000-.111.000000-10 1.00000021 1.000000 ~

-1020.060000-.1 0.6,0000--~01.000000-9 1.00000022 1.000000 -1010.080000-.0 0..00000-39 1.000000-R 1.00000023 1.000000 -1000.100000-69 0.20000 -38 1.000000-~ 1.00000024 1.000000 -99 0.120000-68 0.~~0000-3. 1.000000-6 1.00000025 1.000000 -98 0.1.10000-6. 0..60000-36I1.000000-.~ 1.00000026 1.000000 -9 0.160000-66 0. ; -3.51.000000-.1 1.0000002 1.000000 ~ 80000 ~ ~

-96 0.130000-65 O.t00000-:3.11.000000-3 1.000000-?a1.000000 ~

-9.50.200000-6.10.820000.33 1.000000-'? 1.000000'?91.000000 ~

-9-10.2'?0000-63 0.x.10000-.3Z1.000000-1 1.00000030 1.000000 I, ~

-93 0.2-10000-62 0.?60000-31 1.0000000 1.00000031 1-000000 -92 0.260000-61 0.8.80000-30 1.0000001 1.00000032 1.000000 -91 0.280000-60 0.900000-'?91.0000002 1.00000033 1.000000 -90 0.300000-.590.920000-'231.0000003 1.0000003.11.000000 -39 0.320000-.580.9-10000-~?~1.000000.1 1.0000003.51.000000 -88 0.3-10000-.i 0.960000- 1.000000.~ 1.00000036 1.000000 ~ l6 -87 0.360000-.560.980000-'?~1.0000006 1.0000003 1.000000 ~

-86 0.380000-.i31.000000-2~ 1.000000a 1.00000038 1.000000 -g.50..100000-:W 1.000000-'?31.0000008 1.00000039 1.000000 -8-10.420000-53 1.000000-Z'?1.0000009 1.000000.101.000000 -83 0..140000-.521.000000-21 1.00000010 1.000000-111.000000 -R2 0.460000-.511.000000-'?01.00000011 1.00000042 1.000000 -81 0..180000-SO 1.000000-19 1.00000012 1.00000043 1.000000 -80 0.500000--191.000000-18 1.00000013 1.000000.1-11.000000 -.9 0..520000--181.000000-1. 1.00000014 1.000000-151.000000 - 0..5.10000-4 1.000000-16 1.00000015 1.000000-161.000000 ~ ~

-ii 0.560000-46 1.000000-13 1.00000016 1.00000047 1.000000 - 0.580000-45 1.000000-1~ 1.0000001 1.000000-t81.000000 ~ ~

-~i 0.600000-44 1.000000-1:31.00000013 1.000000.191.000000 SUBSTITUTE SHEET

pCTI US91109135 n n wslm , '~'S1m JO I .OOOOOOr~ 0..18OOOO

51 ! .00000082 0..:60000 52 1.00000033 0..1-10000 i3 1.00000084 0.-120000 i.l L .OOOOOO85 O.-IOOOOO

i5 1.00000086 0.380000 .i6 0.9800008 0.360000 i 0.96000088 0.340000 58 0.9-1000089 0.320000 59 0.92000090 0.300000 60 0.90000091 0.280000 61 O.~.g000092 0.260000 ~

~ 0.6000093 0.10000 63 0.8-1000094 0.220000 6-1 0.82000095 0.200000 65 0.30000096 0.180000 66 0. ; 97 0.160000 :0000 6 0. ~ 98 0.1-10000 ~ 60000 68 0. ~ 99 0.120000 69 0. ~ 100 0.100000 0 0. ~ 101 0.080000 1 0.680000102 0.060000 2 0.660000103 0.040000 7 0.6-10000104 0.020000 7 0.620000105 0.000000 -!

7 0.600000 7 0.580000 7 0.560000 7 0.510000 7 0.520OOO

80 0.500000 SUSSTtTUTE SHEET

PCi'l US91 /09135 Speech .analysis . Pitch Estimation . V/L; V Determination - Spectral .amplitude Estimation Parameter Encoding - Fundamental Frequency Encoding Encoder - V/UV Decision Encoding - Spectral Amplitude Encoding FEC EncodinE
- Error Correction Encoding . Bit Interleaving Channel FEC De~codin~
- Bit De~Interieaving - Error Correction Decoding - Adaptive Smooching parameter Decoding - Fundamental Frequency Decoding Decoder - V/UV Decision Decoding - Spectra) Amplitude Decoding Syegth Synth - Specirai Amplitude Enhancement - Lnvoiced Speech Synthesis - Voiced Speech Synthesis Flow Chart 1: II~iBE Voice Coder SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ (~ ~ ~ PCT/US91/09135 Initial Pitch Estimation Compute E(P) via Equations (5) - (9) Look-Back Pitch Tracking Compute CEB(PB) and PB
via Equations ! 10) - ( 12) Look-Ahead Pitch Tracking, Compute CEF(PF) and PF
via Equations (13) - (20) True eEB(PB) <_ .:gig False CEB(PB) S CEF(PF) Fatse Pt = PF I I Pi = PB
END
Flow Chart 2: Initial Pitch Estimation SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~, ~ ~'~ ~ ~ PCT/US91/09135 Look-Back Pitch Tracking I P=21 I
True P<.8P.1 P=P+.5 False PB = P
.. True -E(P) < E (Ps) PB = P
False l P=P +.5 I
True P < 1 _2p.1 and P S 114 False CEs(P8) = E(Ps) + E_~(P.~) + E.2(P_2) END
Flow Chart 3: Look-Back Pitch Tracking SUBSTITUTE SHEET

Flow Chart 4: Look-Ahead Pitch Tracking (1 of 3) SUBSTITUTE SHEET
(a) (b) (c) {d) WO 92/10830 ~ q ~ ~ ~ ~ PCl"/US91/0913~
(e) Flow Chart 4: Look-Ahead Pitch Tracking {2 of 3) SUBSTITUTE SHEET
(a) (b) (c) WO 92/10830 ~ ~ ~ ~ ~ ~ J PCT/US91/09135 (e) 150 True -~< 21 n=n- 1 n False P _- 1 ~.+ 1 2 ~ 2 Tru a n<2 False ' True CEF(P) <_ .85 and CEF(P) S 1.7 CEF(Po)~-~-False True CEF(P) <_ .4 and CEF(P) S 3.5 CEF(Po) ~---False CEF(P) S .OS True pF = P
False n=n- 1 I
END
Flow Chart 4: Look-Ahead Pitch Tracking (3 of 3) SUBSTITUTE SHEET

WO 92/10830 '~ ~ ~ ~ ~t ~ ~ PCT/US91/09135 t~'/UV Determination Compute K
via Equation (34) Compute ~o via Equation (38) Update Cavg , ~m,uc , ,min via Equations (39) - (43) Compute ~1 (~,p,S",,g,~max~~min) via Equation (44) k=1 Compute Dk via Equation (35) Compute AS(k,w~) via Equation (37) (a) (b) Flow Chart ~: V/UV Determination (1 of 2) SUBSTITUTE SHEET

(a) (b) True vk = 1 ' Dk < AS(k.ci~~) False vk = 0 I k=k+ 1 I
k < K Tie False Compute DK
I via Equation (36) I
..
Compute 8S(K,ci~o) via Equation (37) True DK < g~(K,wo) K -False I VY ~ O I
END
>~ low Chart S: V/UV Determination (2 of 2) SUBSTITUTE SHEET

WO 92/ 10830 ~ ~ ~ ~ ~ ~ PCT/US91 /09135 Unvoiced Speech Synthesis Compute u(n) via Equation (85) Compute Uw(m) via Equation (86) 1=1 Compute a~, bl via Equations (90) - (91) Compute U,.(m) Tcve 1'th spectral magnitude is voiced via Equation (87) , False Compute Uw(m) via Equations (88) - (89) (a) (b) Flow Chart 6: Unvoiced Speech Synthesis (1 of 2) Sl..1l3gTiTUTE SHEET

WO 92/10830 ~ ~ ~ L~ ~ 5 PCT/US91/09135 (a) (b) 1=l+1 I
True E'alse Compute UW(m) via Equation (92) Compute u"(n) via Equation (93) Compute s"~ (n) via Equation (94) END
Elow Chart 6: Unvoiced Speech Synthesis (2 of 2) SUBSTITUTE S~~ET

WO 92/10830 ~ ~ ~ ~ ~ ~ ~ PCT/US9l/09135 Voiced Speech Synthesis Update yy via Equation ( 107) for 1 S 1 <_ 51 l=1 Compute ~~
via Equations (108) - (109) Compute s~.~(n) -rye 1'th spectral magnitude is \
unvoiced in previous frame and via Equation (98) unvoiced in current frame False Compute s',,,i(n) TNe 1'th spectral magnitude is ' voiced in previous frame and via Equation (99) unvoiced in current frame .
False (a) (b) (c) Flow Chart 7: Voiced Speech Synthesis (1 of 2) SUBSTITUTE SHEET

(a) (b) (c) Compute s'~,yn) T,-ue 1'th spectral amplitude is \
unvoiced in previous frame and via Equation ( 100) voiced in current frame False Compute s~,i(n) '1',-ue via Equation ( 101 ) ~~0(0) ' wo(-1 )i ~ .1 wo(0) False Compute s,,,i(n) via Equations (102) - (106) I 1=1+1 I
True t S max ~L(-1),L(0)~,--False Compute s"(n) via Equation (95) END
Flow Chart 7: Voiced Speech Synthesis (2 of 2) SUBSTITUTE SHEET

~ Spectral Amplitude Enhancement ~
Compute R~, R, via Equations (7.~) - (75) I=1 Compute Wi via Equation (76) Vtodiify 1~I~
via Equation (77) I 1=1+1 I
True False Update SE
via Equation (78) END
Flow Chart 8: Spectral Amplitude Enhancement SUSSTlTUTE SHEET

WO 92/10830 ~ ~ ~ ~ ~ PCT/US91/09135 Adaptive Smoothing eT = # of errors detected in c~
True one or more errors _ '7 - ET detected by parity checks in u~
False ET = ~T + # of errors detected in c'i ~T = ET + # of errors detected in c2 ET = ET + # of errors detected in c 3 ET = ~T + # of errors detected in c4 Er = ET + # of errors detected in c5 ET = ~T + # of errors detected in c,~
Update eR
via Equation (79) Compute MT
via Equation (80) (a) Flow Chart 9: Adaptive Smoothing (1 of 3) SUBSTITUTE SHEET

WO 92/10830 ~ ~ ~ ~ ~ r ~ PCi'1US91/09135 (a) 1=1 True yh > 1~T
declare I'th spectral amplitude voiced False I 1=I+1 I
False True ISL
True ET > g and ER(0) <_ .02 True False ER(0) > .1 False (d) (c) Flow Chart 9: adaptive Smoothing (2 of 3) suBSrrTU~ sv~~-r (b) (c) True eT > 12 False True one or more errors detected by parity checks in u~ , False True 192 5 bo <_ 199 False True 204 <_ bo <_ 255 False Perform a frame repeat via Equations (81 ) - (84) END
Flow Chart 9: Adaptive Smoothing (3 of 3) SUBSTITUTE S6-1~~

FEC Encoding Construct up Golay Encode : up -~ cp Construct a 1 Golay Encode : a ~ -~ c ~
Construct u2 Hamming Encode : u2 --> c2 Construct u3 Hamming Encode : u3 --~ c3 (a) Flow Chart 10: FEC Encoding (1 of 2) SUBSTITUTE SHEET

WO 92/ 10830 ~ ~ ~ ~ ~ ~ ~ PCT/US91 /09135 (a) Construct u4 Hamming Encode : u4 -~ c.~ I
Construct u5 Hamming Encode : us -~ cs Construct u6 Hamming Encode : u6 -~ c6 I
Construct u~
c~ = u~
Interframe Interleaving END
Flow Chart 10: FEC Encoding (2 of 2) SUBSTITUTE SHEET

~~~~~~a WO 92/10830 PCTlUS9ll09135 Pitch Refinement P=P1- 1.125 E mi n - ao i wo=~ i Compute S"(m, wo) for SO S m <_ .~ - .S
L~ J25--~J
wo 2II
via Equations (2~ - (28) Compute ER(wo) via Equation (24) E~_ = Elz(wo) True ER(wo) < E
w~ = w~ ~
False (a) (b) (c) Flow Chart 11: Pitch Refinement (1 of Z) SUBSTITUTE SHEET

(a) (b) I P=P +.25 I
True P<_PI+ 1.125' False Compute L
via Equation (31) Compute ai for 1 Sl <_L
via Equation (32) ..
Compute b~
for 1 <_1 <_L
via Equation (33) END
Flow Chart 11: Pitch Refinement (2 of 2) SUBSTITUTE SHEET

pCTlUS91l09135 References (1) L. B. ~lmeida and F. ~I. Silva, "Variable Frequency Synthesis: .W Improved Harmonic Coding Scheme," Proc. ICASSP 84, San Diego, C.~, pp. 289-292, March 1984.
(2) ti. S. Brandstein et. al., ".a Real-Time Implementation of the Improved VIBE Speech Coder," Proc. IC.-1SSP 90, Albuquerque, '.Vi., pp. 5-8, .~pril 1990.
(3J J. P. Campbell et. al., "The new .1800 bps Voice Coding Standard," P~c.
.Mil. Speech Tech. 89, Washington D.C., pp. 64- ~ 0, ~ov. 1989.
(4) B. Vital et. al., Editors, Advances in Speech Coding, pp. 215-224, Kluwer .academic Publishers, 1991.
(.S) D. W. Griffin and J. S. Lim. "Lfultiband Excitation Vocoder," IEEE
Transactions on .-1SSP, Vol. 36, ''o. 8, .august 1988.
(6J D. W. Griffin and J. S. Lim, "Signal Estimation From Modified Short-Time Fourier Transform," IEEE Transactions an .-1SSP, Vol. 32. ~o. 2, pp. 236-243, April 1984.
(7) J. C. Hardwick and J. S. Lim. 'W 4800 bps Improved l~iulti-Band Excitation Speech Coder,'' Pros. of IEEE Workshop on Speech Coding for Tele., Vancouver, B.C., Canada, Sept 5-8, 1989, (8) J. C. Hardwick, "~ 4.8 Kbps Viulti-Band Excitation Speech Coder," S.;~l.
Thesis, E.E.C.S Department, ~I.LT., :stay 1988.
(9) N. Jayant and P. doll, Digital Coding of Waveforms, Prentice-Hall, 1984.
(10) A. Levesque and ~. l~iichelson, Error-Control Techniques /or Digital Communication, Wiley, 1985.
(11J Lin and Costello, Error Control Coding: Fundamentals and Applications, Prentice-Hall, 1983.
(12J l~takhoul et. al., "Vector Quantization in Speech Coding," Pmc. of the IEEE, pp.
1551-1588, 1985.
(13) R. VicAulay and T. Quatieri, "Speech W alysis-Synthesis Based on a Sinusoidal Rep-resentation," IEEE T~'ansactions on ASSP, Vol. ASSP-34, ~lo. 4, aug. 1986.
(14J W. Press et, al., r1%umerical Recipes in C, Cambridge University Press, 1988.
(15) A. Oppenheim and R. Scha.fer, Discrete Time Signal Processing, Prentice-Hall, 1989.
SUBSTITUTE SHEET

Claims (30)

CLAIMS:

1. A method of encoding speech wherein the speech is broken into segments, each of said segments representing one of a succession of time intervals and having a spectrum of frequencies, and for each segment the spectrum of frequencies is sampled at a set of frequencies to form a set of actual spectral amplitudes, with the frequencies at which the spectrum of frequencies is sampled generally differing from one segment to the next, and wherein the spectral amplitudes for at least one previous segment are used to produce a set of predicted spectral amplitudes for a current segment, and wherein a set of prediction residuals for the current segment based on a difference between the actual spectral amplitudes for the current segment and the predicted spectral amplitudes for a current segment are used in subsequent encoding, characterized in that the prediction residuals for a segment are grouped into blocks, the prediction residuals within each block are determined, the averages of each of the blocks are grouped into a prediction residual block average (PRBA) vector, and the PRBA vector is encoded.

2. The method of claim 1 wherein there are a predetermined number of blocks, with the number of blocks being independent of the number of prediction residuals grouped into particular blocks.

3. The method of claim 2 wherein the predicted spectral amplitudes for the current segment are based at least in part on interpolating the spectral amplitudes of a previous segment to estimate the spectral amplitudes in the previous segment at the frequencies of the current segment.

166a

4. The method of claim 1 wherein the difference between the actual spectral amplitudes for the current segment and the predicted spectral amplitudes for the current segment is formed by subtracting a fraction of the predicted spectral amplitudes from the actual spectral amplitudes.

5. The method of claim 1 wherein the spectral amplitudes are obtained using a Multiband Excitation speech model.

6. The method of claim 1 wherein only spectral amplitudes from the most recent previous segment are used in forming the predicted spectral amplitudes of the current segment.

7. The method of claim 1 wherein said spectrum comprises a fundamental frequency and the set of frequencies for a given segment are multiples of the fundamental frequency of the segment.

8. The method of claim 2 or 3 wherein the number of blocks is equal to six (6).

9. The method of claim 2 or 3 wherein the number of prediction residuals in a lower frequency block is not larger than the number of prediction residuals in a higher frequency block.

10. The method of claim 8 wherein the number of prediction residuals in a lower frequency block is not larger than the number of prediction residuals in a higher frequency block.

11. The method of claim 10 wherein the difference between the number of elements in the highest frequency block and the number of elements in the lowest frequency block is less than or equal to one.

12. The method of claim 1, 2 or 3 wherein said average is computed by adding the prediction residuals within the block and dividing by the number of prediction residuals within that block.

13. The method of claim 12 wherein said average is obtained by computing a Discrete Cosine Transform (DCT) of the spectral amplitude prediction residuals within a block and using the first coefficient of the DCT as the average.

14. The method of claim 1, 2 or 3 wherein encoding the PRBA vector comprises vector quantizing the PRBA vector.

15. The method of claim 14 wherein said vector quantization is performed using a method comprising the steps of:
determining an average of the PRBA vector;
quantizing said average using scalar quantization;
subtracting said average from the PRBA vector to form a zero-mean PRBA vector; and quantizing said zero-mean PRBA vector using vector quantization with a zero-mean code-book.

16. The method of claim 1 wherein the speech is encoded using a speech model characterized by model parameters, wherein the speech is broken into time segments and for each segment model parameters quantized, and at least some of the quantized model parameters are coded using error correction coding, with at least two types of error correction coding being used to code the quantized model parameters, including a first type of coding, which adds a greater number of additional bits than a second type of coding, being used for a first group of quantized model parameters that are more sensitive to bit errors than a second group of quantized model parameters.

17. The method of claim 16 wherein the different types of error correction coding including Golay codes and Hamming codes.

18. The method of claim 1 wherein:
for each segment model parameters are quantized, at least some of the quantized model parameters are coded using error correction coding, speech is synthesized from the decoded quantized model parameters, the error correction coding is used in synthesis to estimate an error rate, and one or more model parameters from a previous segment are repeated in a current segment when the error rate for the parameter exceeds a predetermined level.

19. The method of claim 16, 17 or 18 wherein the quantized model parameters are those associated with a Multi-Band Excitation (MBE) speech coder or Improved Multi-Band Excitation (IMBE) speech coder.

20. The method of claim 16 or 17 wherein error rates are estimated using the error correction codes.

21. The method of claim 20 wherein one or more model parameters are smoothed across a plurality of segments based on estimated error rate.

22. The method of claim 21 wherein the model parameters smoothed include voiced/unvoiced decisions.

23. The method of claim 21 wherein the model parameters smoothed include parameters for a Multi-Band Excitation (MBE) speech coder or Improved MultiBand Excitation (IMBE) speech coder.

24. The method of claim 23 wherein the value of one or more model parameters in a previous segment are repeated in a current segment when the estimated error rate for the parameters exceeds a predetermined level.

25. The method of claim 1 wherein:
frequency domain representations of a segment are determined to provide a spectral envelope of the segment, speech is synthesized from an enhanced spectral envelope, a smoothed spectral envelope of the segment is generated by smoothing the spectral envelop, and an enhanced spectral envelope is generated by increasing some frequency regions of the spectral envelope for which the spectral envelope has greater amplitude than the smoothed envelope and decreasing some frequency regions for which the spectral envelope has lesser amplitude than the smoothed envelope.

26. The method of claim 25 wherein the frequency domain representation of the spectral envelope is the set of spectral amplitude parameters of a Multi-Band Excitation (MBE) speech coder or Improved Multi-Band (IMBE) speech coder.

27. The method of claim 25 or 26 wherein the smoothed spectral envelope is generated by estimating a low-order model from the spectral envelope.

28. The method of claim 27 wherein the low-order model is an all-pole model.

29. The method of claim 1, 2 or 3 wherein the PRBA
vector is encoded using a linear transform on the PRBA
vector and scalar quantizing transform coefficients.

30. The method of claim 29 wherein said linear transform comprises a Discrete Cosine Transform.